feat(api): count the quota by the CDN's client IP and refuse direct calls

Two optional settings for deployments behind a CDN such as Cloudflare.

CLIENT_IP_HEADER names the header that holds the visitor's real IP
(cf-connecting-ip on Cloudflare). The per-IP daily quota used the first
X-Forwarded-For entry, which visitors can set to anything, so a made-up
IP on every request got a fresh quota.

ORIGIN_SECRET makes proxy.ts refuse /api requests whose X-Origin-Secret
header does not match. The CDN adds the header, so a call that skips the
CDN, and could fake the IP header, gets 403. Pages are not checked, which
keeps health checks on / working.

Both are unset by default, and nothing changes then.

.github/workflows/auto-format.yml

@@ -25,7 +25,7 @@ jobs:
       - name: Run Biome format
         # Pin to the version in package.json so CI matches local/pre-commit
         # (npx @latest drifts — e.g. 2.5.0 broke this job on unrelated PRs).
-        run: npx @biomejs/biome@2.4.13 check --write --no-errors-on-unmatched .
+        run: npx @biomejs/biome@2.5.7 check --write --no-errors-on-unmatched .
 
       - name: Check for changes
         id: changes

.github/workflows/publish-mcp.yml

@@ -62,6 +62,10 @@ jobs:
         if: steps.version.outputs.publish == 'true'
         run: npm test
 
+      - name: Build and check package contents
+        if: steps.version.outputs.publish == 'true'
+        run: npm run build && npm run check-package
+
       - name: Publish to npm
         if: steps.version.outputs.publish == 'true'
         run: npm publish

.github/workflows/test.yml

@@ -38,6 +38,10 @@ jobs:
       - name: Run MCP server unit tests
         run: npm --prefix packages/mcp-server test
 
+      # Tests run from src/, so check the built npm package separately
+      - name: Build MCP server and check package contents
+        run: npm --prefix packages/mcp-server run build && npm --prefix packages/mcp-server run check-package
+
   e2e:
     name: E2E Tests
     runs-on: ubuntu-latest

.gitignore

@@ -79,5 +79,3 @@ ai-models.json
 
 # admin panel settings (contains secrets)
 data/
-# ...but the diagram engine's stencil catalog is generated source, not settings
-!lib/diagram-engine/data/

NOTICE

@@ -1,52 +0,0 @@
-next-ai-draw-io
-Copyright the next-ai-draw-io contributors
-
-This product includes software developed by third parties, as set out below.
-
---------------------------------------------------------------------------------
-lib/diagram-engine/ — layout and rendering
---------------------------------------------------------------------------------
-
-The declarative layout algorithm (bottom-up measure, top-down place, sibling
-size equalisation), the swimlane-pool geometry, and the mxCell/style emission in
-`lib/diagram-engine/layout.ts` and `lib/diagram-engine/render.ts` are derived
-from drawio-ai-kit:
-
-    https://github.com/sparklabx/drawio-ai-kit
-    Copyright (c) sparklabx
-    Licensed under the MIT License
-
-Original to this repository:
-
-  - the XML→tree reverse parser (`parse.ts`), which that project does not have
-  - the style-marker scheme (`markers.ts`) that lets structure survive a
-    round-trip through the draw.io editor
-  - the structural-operations layer (`operations.ts`)
-  - the invisible-container approach that replaces that project's "phantom" nodes
-  - the edge router (`route.ts`)
-  - the graph→layers pass (`graph.ts`): layer assignment, cycle breaking and
-    barycentre crossing reduction, which turn a flat node/arrow list into a
-    flowchart
-  - the sequence-diagram and radial (mind map / org chart) layouts
-
---------------------------------------------------------------------------------
-lib/diagram-engine/data/aws-stencils.json — stencil catalog
---------------------------------------------------------------------------------
-
-A name→style map for the mxgraph.aws4 stencil family, generated from
-drawio-ai-kit's `catalog/aws.json`, which in turn was generated from the draw.io
-shape index published by jgraph:
-
-    https://github.com/jgraph/drawio-mcp
-    Copyright (c) JGraph Ltd
-    Licensed under the Apache License, Version 2.0
-
-The style strings are reproduced verbatim from that index. They reference the
-official AWS Architecture Icons, which are trademarks of Amazon Web Services and
-are NOT covered by this repository's licence. Their use is governed by the AWS
-Architecture Icons terms:
-
-    https://aws.amazon.com/architecture/icons/
-
-The catalog contains stencil *names and style strings* only — it does not embed
-any AWS icon artwork. draw.io supplies the artwork at render time.

README.md

@@ -21,6 +21,17 @@ A Next.js web application that integrates AI capabilities with draw.io diagrams.
 
 > Note: Thanks to <img src="https://raw.githubusercontent.com/DayuanJiang/next-ai-draw-io/main/public/doubao-color.png" alt="" height="20" /> [ByteDance Doubao](https://www.volcengine.com/activity/codingplan?ac=MMAP8JTTCAQ2&rc=Z9Z3LDTJ&utm_campaign=drawio&utm_content=drawio&utm_medium=devrel&utm_source=OWO&utm_term=drawio) sponsorship, the demo site now uses the powerful glm-4.7 model!
 
+<p align="center">
+  <a href="https://www.atlascloud.ai/?utm_source=github&utm_medium=link&utm_campaign=next-ai-draw-io">
+    <picture>
+      <source media="(prefers-color-scheme: dark)" srcset="./public/atlas-cloud-logo-white.svg">
+      <img src="./public/atlas-cloud-logo.svg" alt="Atlas Cloud" width="200">
+    </picture>
+  </a>
+</p>
+
+> 🎁 Thanks to **[Atlas Cloud](https://www.atlascloud.ai/?utm_source=github&utm_medium=link&utm_campaign=next-ai-draw-io)** for sponsoring next-ai-draw-io. Its OpenAI-compatible API gives diagram workflows one provider connection for DeepSeek, Qwen, GLM, Kimi, MiniMax, and more. Budget-friendly access is available through the [Coding Plan](https://www.atlascloud.ai/console/coding-plan).
+
 
 https://github.com/user-attachments/assets/9d60a3e8-4a1c-4b5e-acbb-26af2d3eabd1
 
@@ -99,7 +110,6 @@ Here are some example prompts and their generated diagrams:
 -   **Diagram History**: Comprehensive version control that tracks all changes, allowing you to view and restore previous versions of your diagrams before the AI editing.
 -   **Interactive Chat Interface**: Communicate with AI to refine your diagrams in real-time
 -   **Cloud Architecture Diagram Support**: Specialized support for generating cloud architecture diagrams (AWS, GCP, Azure)
--   **Computed Layout**: For architecture diagrams, flowcharts, swimlane/BPMN diagrams, sequence diagrams, mind maps and org charts, the AI declares only the structure — what contains what, or what points at what — and the app computes every coordinate, size and arrow route. Containers always fit their contents, siblings never overlap, and arrows are routed around the shapes they would otherwise cross. Anything you then move or recolour by hand is read back as part of the diagram, so a later edit does not undo it.
 -   **Animated Connectors**: Create dynamic and animated connectors between diagram elements for better visualization
 
 ## MCP Server
@@ -128,6 +138,13 @@ Then ask Claude to create diagrams:
 
 The diagram appears in your browser in real-time!
 
+The MCP server includes most of the web app's drawing features:
+
+-   The same drawing rules and shape libraries (AWS, Azure, GCP, Kubernetes and more)
+-   A screenshot tool, so the AI can check the rendered diagram and fix it
+-   Version history, multi-page diagrams, and download as `.drawio`, `.png`, `.svg`, or `.drawio.svg`
+-   Auto-save to `~/.next-ai-drawio/`, so you can continue a diagram after a restart
+
 See the [MCP Server README](./packages/mcp-server/README.md) for VS Code, Cursor, and other client configurations.
 
 ## Getting Started
@@ -218,6 +235,7 @@ See the [Next.js deployment documentation](https://nextjs.org/docs/app/building-
 -   ModelScope
 -   SGLang
 -   Vercel AI Gateway
+-   [Atlas Cloud](https://www.atlascloud.ai/?utm_source=github&utm_medium=link&utm_campaign=next-ai-draw-io)
 
 
 All providers except AWS Bedrock and OpenRouter support custom endpoints.
@@ -254,6 +272,8 @@ Diagrams are represented as XML that can be rendered in draw.io. The AI processe
 
 **Special thanks to [ByteDance Doubao](https://www.volcengine.com/activity/codingplan?ac=MMAP8JTTCAQ2&rc=Z9Z3LDTJ&utm_campaign=drawio&utm_content=drawio&utm_medium=devrel&utm_source=OWO&utm_term=drawio) for sponsoring the API token usage of the demo site!** Register on the ARK platform to get 500K free tokens for all models!
 
+**Special thanks to [Atlas Cloud](https://www.atlascloud.ai/?utm_source=github&utm_medium=link&utm_campaign=next-ai-draw-io) for sponsoring next-ai-draw-io and supporting its multi-provider ecosystem!** Try its OpenAI-compatible LLM API through the [Atlas Cloud Coding Plan](https://www.atlascloud.ai/console/coding-plan).
+
 If you find this project useful, please consider [sponsoring](https://github.com/sponsors/DayuanJiang) to help me host the live demo site!
 
 For support or inquiries, please open an issue on the GitHub repository or contact the maintainer at:

app/[lang]/about/cn/page.tsx

@@ -160,11 +160,6 @@ export default function AboutCN() {
                             <strong>AWS架构图支持</strong>
                             ：专门支持生成AWS架构图
                         </li>
-                        <li>
-                            <strong>自动计算布局</strong>
-                            ：画架构图、流程图、泳道图、时序图、思维导图和组织架构图时，AI
-                            只描述结构，所有坐标、尺寸和连线路径都由程序计算——容器一定装得下里面的内容，图形不会重叠，连线会绕开本来会穿过的图形
-                        </li>
                         <li>
                             <strong>动画连接器</strong>
                             ：在图表元素之间创建动态动画连接器，实现更好的可视化效果

app/[lang]/about/ja/page.tsx

@@ -153,11 +153,6 @@ export default function AboutJA() {
                             </strong>
                             ：AWSアーキテクチャダイアグラムの生成を専門的にサポート
                         </li>
-                        <li>
-                            <strong>レイアウトの自動計算</strong>
-                            ：アーキテクチャ図、フローチャート、スイムレーン図、シーケンス図、マインドマップ、組織図では、AI
-                            は構造だけを指定し、座標・サイズ・矢印の経路はすべてアプリが計算します——コンテナは必ず中身が収まり、図形は重ならず、矢印は本来通り抜けてしまう図形を避けて引かれます
-                        </li>
                         <li>
                             <strong>アニメーションコネクタ</strong>
                             ：より良い可視化のためにダイアグラム要素間に動的でアニメーション化されたコネクタを作成

app/[lang]/about/page.tsx

@@ -165,15 +165,6 @@ export default function About() {
                             Specialized support for generating AWS architecture
                             diagrams
                         </li>
-                        <li>
-                            <strong>Computed Layout</strong>: For architecture
-                            diagrams, flowcharts, swimlane diagrams, sequence
-                            diagrams, mind maps and org charts, the AI declares
-                            only the structure and the app computes every
-                            coordinate, size and arrow route — so containers fit
-                            their contents, shapes never overlap, and arrows are
-                            routed around what they would otherwise cross
-                        </li>
                         <li>
                             <strong>Animated Connectors</strong>: Create dynamic
                             and animated connectors between diagram elements for

app/[lang]/admin/models-section.tsx

@@ -33,8 +33,10 @@ import {
 import { Switch } from "@/components/ui/switch"
 import { useDictionary } from "@/hooks/use-dictionary"
 import { formatMessage } from "@/lib/i18n/utils"
+import { STORAGE_KEYS } from "@/lib/storage"
 import {
     FIXED_CRED_PROVIDERS,
+    generateId,
     PROVIDER_INFO,
     type ProviderName,
     SUGGESTED_MODELS,
@@ -87,6 +89,11 @@ function ProviderDetail({
         try {
             const data = await adminFetch("/api/admin/test-model", password, {
                 method: "POST",
+                // EdgeOne's function also checks the access code
+                headers: {
+                    "x-access-code":
+                        localStorage.getItem(STORAGE_KEYS.accessCode) || "",
+                },
                 body: JSON.stringify({ provider, modelId }),
             })
             setTestResults((prev) => ({
@@ -225,6 +232,7 @@ function ProviderDetail({
                         </Button>
                         {suggestions.length > 0 && (
                             <Select
+                                value=""
                                 disabled={disabled}
                                 onValueChange={(v) => addModel(v)}
                             >
@@ -390,12 +398,14 @@ function ProviderDetail({
 export function ModelsSection({
     providers,
     envProviders,
+    envHasDefaultModel,
     disabled,
     password,
     onChange,
 }: {
     providers: AdminProvider[]
     envProviders: EnvProvider[]
+    envHasDefaultModel: boolean
     disabled: boolean
     password: string
     onChange: (providers: AdminProvider[]) => void
@@ -409,10 +419,16 @@ export function ModelsSection({
 
     const addProvider = (provider: ProviderName) => {
         const newProvider: AdminProvider = {
-            id: crypto.randomUUID(),
+            // generateId works over plain HTTP; crypto.randomUUID needs HTTPS
+            id: generateId(),
             provider,
             models: [],
-            isDefault: providers.length === 0,
+            // Only the very first provider becomes the default, and only when
+            // the env config has no default that it would replace on save
+            isDefault:
+                providers.length === 0 &&
+                !envProviders.some((p) => p.isDefault) &&
+                !envHasDefaultModel,
         }
         onChange([...providers, newProvider])
         setSelectedId(newProvider.id)
@@ -496,7 +512,9 @@ export function ModelsSection({
                     ))}
                 </div>
                 <div className="border-t p-2">
+                    {/* Always empty so picking the same type again still fires */}
                     <Select
+                        value=""
                         disabled={disabled}
                         onValueChange={(v) => addProvider(v as ProviderName)}
                     >

app/[lang]/admin/page.tsx

@@ -37,6 +37,19 @@ import { SettingField } from "./setting-field"
 
 const NAV_GROUP_IDS = ["models", ...SETTING_GROUPS.map((g) => g.id)]
 
+// For each toggleable group, whether any of its settings has a value (from
+// the settings file or the environment)
+function groupsWithValues(map: SettingsMap): Record<string, boolean> {
+    const result: Record<string, boolean> = {}
+    for (const group of SETTING_GROUPS) {
+        if (!group.toggleable) continue
+        result[group.id] = !!SETTINGS_BY_GROUP.get(group.id)?.some(
+            (d) => map[d.key]?.source !== "default",
+        )
+    }
+    return result
+}
+
 export default function AdminPage() {
     const dict = useDictionary()
     // Localized group title/description, keyed by group id
@@ -62,6 +75,8 @@ export default function AdminPage() {
     // Models section state
     const [providers, setProviders] = useState<AdminProvider[]>([])
     const [envProviders, setEnvProviders] = useState<EnvProvider[]>([])
+    // Whether .env itself sets AI_MODEL (a default the panel would override)
+    const [envHasDefaultModel, setEnvHasDefaultModel] = useState(false)
     const [savedProviders, setSavedProviders] = useState<string>("[]")
     const providersDirty = JSON.stringify(providers) !== savedProviders
 
@@ -88,15 +103,13 @@ export default function AdminPage() {
             const map: SettingsMap = {}
             for (const s of data.settings) map[s.key] = s
             setSettings(map)
-            // Seed each toggle once from whether the group has configured
-            // values; don't stomp a user's explicit toggle on later saves
+            // A group stays on while it still has values (e.g. from env vars
+            // that saving can't remove); a user's explicit "on" for a group
+            // with no values yet is kept across saves
             setEnabledGroups((prev) => {
-                const next = { ...prev }
-                for (const group of SETTING_GROUPS) {
-                    if (!group.toggleable || group.id in next) continue
-                    next[group.id] = !!SETTINGS_BY_GROUP.get(group.id)?.some(
-                        (d) => map[d.key]?.source !== "default",
-                    )
+                const next = groupsWithValues(map)
+                for (const id of Object.keys(next)) {
+                    next[id] = next[id] || !!prev[id]
                 }
                 return next
             })
@@ -108,10 +121,12 @@ export default function AdminPage() {
         (data: {
             providers: AdminProvider[]
             envProviders?: EnvProvider[]
+            envHasDefaultModel?: boolean
         }) => {
             setProviders(data.providers)
             setSavedProviders(JSON.stringify(data.providers))
             setEnvProviders(data.envProviders ?? [])
+            setEnvHasDefaultModel(!!data.envHasDefaultModel)
         },
         [],
     )
@@ -181,8 +196,9 @@ export default function AdminPage() {
         return () => observer.disconnect()
     }, [authedPassword])
 
+    // value undefined drops the pending change (back to the saved value)
     const handleChange = useCallback(
-        (key: string, value: string | null) => {
+        (key: string, value: string | null | undefined) => {
             setSaveMessage(null)
             setErrors((prev) => {
                 if (!(key in prev)) return prev
@@ -201,7 +217,7 @@ export default function AdminPage() {
                     value === "" &&
                     (!state || state.source !== "file") &&
                     !isSecretValue(state?.value)
-                if (isRevert || isNoop) {
+                if (value === undefined || isRevert || isNoop) {
                     const next = { ...prev }
                     delete next[key]
                     return next
@@ -225,9 +241,10 @@ export default function AdminPage() {
                 const next = { ...prev }
                 for (const key of keys) {
                     if (!enabled) {
-                        // Stage deletion only for values currently set
-                        if (settings[key]?.source !== "default")
-                            next[key] = null
+                        // Stage deletion of saved values; drop unsaved input
+                        if (settings[key]?.source === "default")
+                            delete next[key]
+                        else next[key] = null
                     } else if (next[key] === null) {
                         delete next[key]
                     }
@@ -447,6 +464,7 @@ export default function AdminPage() {
                             <ModelsSection
                                 providers={providers}
                                 envProviders={envProviders}
+                                envHasDefaultModel={envHasDefaultModel}
                                 disabled={!writable || saving}
                                 password={authedPassword}
                                 onChange={(next) => {
@@ -462,6 +480,11 @@ export default function AdminPage() {
                         const defs = SETTINGS_BY_GROUP.get(group.id) ?? []
                         const groupOff =
                             group.toggleable && !enabledGroups[group.id]
+                        // Values from env vars can't be removed here, so the
+                        // group can't be turned off from the panel
+                        const envLocked = defs.some(
+                            (d) => settings[d.key]?.source === "env",
+                        )
                         const fieldsDisabled = !writable || saving || !!groupOff
                         const gt = groupText(group.id)
                         const title = gt?.title ?? group.title
@@ -480,6 +503,11 @@ export default function AdminPage() {
                                     </h2>
                                     {group.toggleable && (
                                         <label
+                                            title={
+                                                envLocked
+                                                    ? dict.admin.sourceEnvTitle
+                                                    : undefined
+                                            }
                                             className={cn(
                                                 "flex cursor-pointer items-center gap-2 rounded-full border px-3 py-1.5 text-xs font-medium transition-colors motion-reduce:transition-none",
                                                 enabledGroups[group.id]
@@ -494,7 +522,11 @@ export default function AdminPage() {
                                                 checked={
                                                     !!enabledGroups[group.id]
                                                 }
-                                                disabled={!writable || saving}
+                                                disabled={
+                                                    !writable ||
+                                                    saving ||
+                                                    envLocked
+                                                }
                                                 aria-label={formatMessage(
                                                     dict.admin.enableGroup,
                                                     { group: title },
@@ -579,6 +611,9 @@ export default function AdminPage() {
                                         setPending({})
                                         setErrors({})
                                         setProviders(JSON.parse(savedProviders))
+                                        setEnabledGroups(
+                                            groupsWithValues(settings),
+                                        )
                                     }}
                                 >
                                     {dict.admin.discard}

app/[lang]/admin/setting-field.tsx

@@ -73,8 +73,10 @@ export function SecretInput({
 }) {
     const dict = useDictionary()
     const [show, setShow] = useState(false)
-    // The stored marker as it was at mount, to revert to on empty
-    const [original] = useState(value)
+    // The stored marker to revert to on empty. Refreshed whenever the parent
+    // passes server state (a marker or nothing), e.g. after a save.
+    const [original, setOriginal] = useState(value)
+    if (typeof value !== "string" && value !== original) setOriginal(value)
     const hadStored = isSecretValue(original)
     const text = typeof value === "string" ? value : ""
     const placeholder = isSecretValue(value)
@@ -146,7 +148,8 @@ export function SettingField({
     pendingValue: string | null | undefined
     error?: string
     disabled: boolean
-    onChange: (value: string | null) => void
+    // undefined drops the pending change (back to the saved value)
+    onChange: (value: string | null | undefined) => void
 }) {
     const dict = useDictionary()
     const isDirty = pendingValue !== undefined
@@ -226,16 +229,18 @@ export function SettingField({
         case "secret":
             control = (
                 <div className="w-full max-w-md">
+                    {/* Clearing a saved secret reverts to it; the X button deletes */}
                     <SecretInput
                         id={inputId}
+                        keepOnEmpty={source === "file"}
                         value={
                             isDirty
                                 ? (pendingValue ?? "")
-                                : (secretState ?? currentValue)
+                                : (secretState ?? undefined)
                         }
                         disabled={disabled}
                         onChange={(v) =>
-                            onChange(typeof v === "string" ? v : "")
+                            onChange(typeof v === "string" ? v : undefined)
                         }
                     />
                 </div>

app/[lang]/page.tsx

@@ -37,7 +37,6 @@ export default function Home() {
     )
 
     const chatPanelRef = useRef<ImperativePanelHandle>(null)
-    const isMobileRef = useRef(false)
 
     // Load preferences from localStorage after mount
     useEffect(() => {
@@ -48,7 +47,9 @@ export default function Home() {
             const currentLocale = pathParts[0]
             if (currentLocale !== savedLocale) {
                 pathParts[0] = savedLocale
-                router.replace(`/${pathParts.join("/")}`)
+                // Keep the query (e.g. ?session=) and hash
+                const { search, hash } = window.location
+                router.replace(`/${pathParts.join("/")}${search}${hash}`)
                 return // Wait for redirect
             }
         }
@@ -106,27 +107,25 @@ export default function Home() {
         resetDrawioReady()
     }
 
-    // Check mobile - reset draw.io before crossing breakpoint
-    const isInitialRenderRef = useRef(true)
+    // Check mobile. No panel is remounted when crossing the breakpoint, so
+    // the draw.io ready state and the chat's turn stay as they are.
     useEffect(() => {
         const checkMobile = () => {
-            const newIsMobile = window.innerWidth < 768
-            if (
-                !isInitialRenderRef.current &&
-                newIsMobile !== isMobileRef.current
-            ) {
-                setIsDrawioReady(false)
-                resetDrawioReady()
-            }
-            isMobileRef.current = newIsMobile
-            isInitialRenderRef.current = false
-            setIsMobile(newIsMobile)
+            setIsMobile(window.innerWidth < 768)
         }
 
         checkMobile()
         window.addEventListener("resize", checkMobile)
         return () => window.removeEventListener("resize", checkMobile)
-    }, [resetDrawioReady])
+    }, [])
+
+    // Give the chat panel the size of this side of the breakpoint. It is
+    // open on both sides: the mobile panel cannot be collapsed, and one
+    // collapsed on desktop comes back open
+    useEffect(() => {
+        chatPanelRef.current?.resize(isMobile ? 50 : 33)
+        setIsChatVisible(true)
+    }, [isMobile])
 
     const toggleChatPanel = () => {
         const panel = chatPanelRef.current
@@ -193,7 +192,11 @@ export default function Home() {
                                             noExitBtn: true,
                                             dark:
                                                 darkMode || drawioUi === "dark",
-                                            lang: currentLang,
+                                            // draw.io names Traditional Chinese "zh-tw"
+                                            lang:
+                                                currentLang === "zh-Hant"
+                                                    ? "zh-tw"
+                                                    : currentLang,
                                             // Enable offline mode in Electron to disable external service calls
                                             ...(isElectron && {
                                                 offline: true,
@@ -217,7 +220,6 @@ export default function Home() {
 
                 {/* Chat Panel */}
                 <ResizablePanel
-                    key={isMobile ? "mobile" : "desktop"}
                     id="chat-panel"
                     ref={chatPanelRef}
                     defaultSize={isMobile ? 50 : 33}

app/api/admin/providers/route.ts

@@ -7,7 +7,11 @@ import {
     mergeSecrets,
     validateAdminProviders,
 } from "@/lib/admin/providers"
-import { isSettingsWritable, saveSettings } from "@/lib/admin/settings"
+import {
+    getEnvFallback,
+    isSettingsWritable,
+    saveSettings,
+} from "@/lib/admin/settings"
 import { loadEnvServerModelsConfig } from "@/lib/server-model-config"
 
 export const runtime = "nodejs"
@@ -33,6 +37,9 @@ async function payload() {
                 models: p.models,
                 isDefault: !!p.default && !adminHasDefault,
             })) ?? [],
+        // Whether .env sets a default model. getEnvFallback skips the value
+        // the panel overlays onto process.env, so a panel default doesn't count.
+        envHasDefaultModel: !!getEnvFallback("AI_MODEL"),
     }
 }
 

app/api/admin/test-model/route.ts

@@ -5,6 +5,7 @@ import {
     loadAdminProviders,
     mergeSecrets,
 } from "@/lib/admin/providers"
+import { globalBaseUrl } from "@/lib/ai-providers"
 
 export const runtime = "nodejs"
 export const dynamic = "force-dynamic"
@@ -47,14 +48,31 @@ export async function POST(req: Request) {
         sameEndpoint && stored ? [stored] : [],
     )
 
+    const serverUrl = globalBaseUrl(resolved.provider)
     return validateModel(
         new Request(new URL("/api/validate-model", req.url), {
             method: "POST",
-            headers: { "Content-Type": "application/json" },
+            headers: {
+                "Content-Type": "application/json",
+                // Checked again there, in place of an access code
+                "x-admin-password": req.headers.get("x-admin-password") || "",
+                // The EdgeOne function checks the access code and Pages
+                // cookies, and its URL is built from the page's origin
+                "x-access-code": req.headers.get("x-access-code") || "",
+                cookie: req.headers.get("cookie") || "",
+                ...(req.headers.get("origin") && {
+                    origin: req.headers.get("origin") as string,
+                }),
+            },
             body: JSON.stringify({
                 provider: resolved.provider,
                 apiKey: resolved.apiKey,
-                baseUrl: resolved.baseUrl,
+                // Without a URL of its own, chat sends the entry's key to
+                // the server's <P>_BASE_URL: test that endpoint, not
+                // another one. It is the server's own, which chat uses
+                // without the checks for a URL a user typed.
+                baseUrl: resolved.baseUrl || serverUrl,
+                ...(!resolved.baseUrl && serverUrl && { serverBaseUrl: true }),
                 modelId: body.modelId,
                 awsAccessKeyId: resolved.awsAccessKeyId,
                 awsSecretAccessKey: resolved.awsSecretAccessKey,

app/api/aihubmix-models/route.ts

@@ -1,61 +0,0 @@
-import { NextResponse } from "next/server"
-import {
-    AIHUBMIX_MODELS_ENDPOINT,
-    extractAihubmixModelIds,
-} from "@/lib/aihubmix-models"
-import { SUGGESTED_MODELS } from "@/lib/types/model-config"
-
-const SUCCESS_CACHE_CONTROL =
-    "public, max-age=300, s-maxage=3600, stale-while-revalidate=86400"
-
-function fallbackResponse() {
-    return NextResponse.json(
-        {
-            models: SUGGESTED_MODELS.aihubmix || [],
-            source: "fallback",
-        },
-        {
-            headers: {
-                "Cache-Control": "no-store",
-            },
-        },
-    )
-}
-
-export async function GET() {
-    try {
-        const response = await fetch(AIHUBMIX_MODELS_ENDPOINT, {
-            next: { revalidate: 3600 },
-        })
-
-        if (!response.ok) {
-            console.warn(
-                `[aihubmix-models] Failed to fetch models: ${response.status}`,
-            )
-            return fallbackResponse()
-        }
-
-        const payload = await response.json()
-        const models = extractAihubmixModelIds(payload)
-
-        if (models.length === 0) {
-            console.warn("[aihubmix-models] Model list response was empty")
-            return fallbackResponse()
-        }
-
-        return NextResponse.json(
-            {
-                models,
-                source: "aihubmix",
-            },
-            {
-                headers: {
-                    "Cache-Control": SUCCESS_CACHE_CONTROL,
-                },
-            },
-        )
-    } catch (error) {
-        console.warn("[aihubmix-models] Failed to load models:", error)
-        return fallbackResponse()
-    }
-}

app/api/log-save/route.ts

@@ -4,7 +4,7 @@ import { getLangfuseClient } from "@/lib/langfuse"
 
 const saveSchema = z.object({
     filename: z.string().min(1).max(255),
-    format: z.enum(["drawio", "png", "svg"]),
+    format: z.enum(["drawio", "png", "svg", "xmlsvg"]),
     sessionId: z.string().min(1).max(200).optional(),
 })
 

app/api/parse-url/route.ts

@@ -1,9 +1,12 @@
 import { extractFromHtml } from "@extractus/article-extractor"
 import { NextResponse } from "next/server"
 import TurndownService from "turndown"
+import { checkAccessCode, rejectCrossSite } from "@/lib/access-code"
+import { readLimitedBody } from "@/lib/read-limited-body"
 import { isPrivateUrl } from "@/lib/ssrf-protection"
 
 const MAX_CONTENT_LENGTH = 150000 // Match PDF limit
+const MAX_RESPONSE_BYTES = 5 * 1024 * 1024
 const EXTRACT_TIMEOUT_MS = 15000
 const USER_AGENT = "Mozilla/5.0 (compatible; NextAIDrawio/1.0)"
 
@@ -33,6 +36,11 @@ function detectCharset(
 }
 
 export async function POST(req: Request) {
+    const crossSite = rejectCrossSite(req)
+    if (crossSite) return crossSite
+    const accessError = checkAccessCode(req)
+    if (accessError) return accessError
+
     try {
         const { url } = await req.json()
 
@@ -97,7 +105,15 @@ export async function POST(req: Request) {
                 )
             }
 
-            const buffer = await response.arrayBuffer()
+            const buffer = await readLimitedBody(response, MAX_RESPONSE_BYTES)
+            if (!buffer) {
+                return NextResponse.json(
+                    {
+                        error: `Page exceeds the ${MAX_RESPONSE_BYTES / 1024 / 1024} MB download limit`,
+                    },
+                    { status: 413 },
+                )
+            }
             const charset = detectCharset(contentType, buffer)
             html = new TextDecoder(charset).decode(buffer)
         } catch (err: any) {
@@ -115,6 +131,9 @@ export async function POST(req: Request) {
             )
         } finally {
             clearTimeout(timeoutId)
+            // Ends a download left unread (too large, PDF, error status);
+            // a body already read is not affected
+            controller.abort()
         }
 
         // extractFromHtml throws (not returns null) on empty/non-HTML bodies,

app/api/provider-models/route.ts

@@ -0,0 +1,79 @@
+import { NextResponse } from "next/server"
+import { checkAccessCode, rejectCrossSite } from "@/lib/access-code"
+import { classifyLLMError } from "@/lib/llm-errors"
+import {
+    canListModels,
+    listProviderModels,
+    ModelListError,
+} from "@/lib/provider-models"
+import {
+    allowPrivateUrls,
+    isPrivateUrl,
+    RedirectRefusedError,
+    redirectGuardedFetch,
+} from "@/lib/ssrf-protection"
+import type { ProviderName } from "@/lib/types/model-config"
+
+export const runtime = "nodejs"
+
+// Public lists need no key
+const NO_KEY_NEEDED = new Set<ProviderName>([
+    "ollama",
+    "openrouter",
+    "aihubmix",
+])
+
+/**
+ * The models a provider offers, for the "Fetch models" button in model
+ * settings. Answers { models: null } for providers that cannot list them,
+ * so the dialog keeps its suggested models.
+ */
+export async function POST(req: Request) {
+    const crossSite = rejectCrossSite(req)
+    if (crossSite) return crossSite
+    // Sends requests to a URL the client chose, so require the access code
+    const accessError = checkAccessCode(req)
+    if (accessError) return accessError
+
+    const { provider, apiKey, baseUrl } = (await req.json()) as {
+        provider: ProviderName
+        apiKey?: string
+        baseUrl?: string
+    }
+    if (!canListModels(provider)) {
+        return NextResponse.json({ models: null })
+    }
+    // SECURITY: Block SSRF attacks via custom baseUrl
+    if (baseUrl && !allowPrivateUrls() && (await isPrivateUrl(baseUrl))) {
+        return NextResponse.json({ error: "Invalid base URL" }, { status: 400 })
+    }
+    if (!apiKey && !NO_KEY_NEEDED.has(provider)) {
+        return NextResponse.json(
+            { error: "API key is required" },
+            { status: 400 },
+        )
+    }
+
+    try {
+        const models = await listProviderModels(
+            provider,
+            { apiKey, baseUrl },
+            (baseUrl && redirectGuardedFetch()) || fetch,
+        )
+        return NextResponse.json({ models })
+    } catch (error) {
+        console.warn("[provider-models] Listing failed:", error)
+        // Only our own explanations go back: the URL may be an internal
+        // address, whose answer or host names must not reach the caller.
+        // The Gateway SDK wraps them, keeping ours as the cause.
+        const isOwn = (e: unknown): e is Error =>
+            e instanceof ModelListError || e instanceof RedirectRefusedError
+        const cause = (error as { cause?: unknown })?.cause
+        const own = isOwn(error) ? error : isOwn(cause) ? cause : null
+        const { code } = classifyLLMError(own ?? error)
+        return NextResponse.json({
+            code,
+            error: own?.message ?? "The model list request failed.",
+        })
+    }
+}

app/api/validate-diagram/route.ts

@@ -3,8 +3,15 @@
  * Accepts a PNG image and streams validation results using useObject-compatible format.
  */
 
-import { streamObject } from "ai"
+import { Output, streamText } from "ai"
+import { checkAccessCode, rejectCrossSite } from "@/lib/access-code"
 import { getValidationModel } from "@/lib/ai-providers"
+import {
+    checkAndIncrementRequest,
+    isQuotaEnabled,
+    recordTokenUsage,
+} from "@/lib/dynamo-quota-manager"
+import { getUserIdFromRequest } from "@/lib/user-id"
 import { VALIDATION_SYSTEM_PROMPT } from "@/lib/validation-prompts"
 import {
     type ValidationResult,
@@ -13,6 +20,9 @@ import {
 
 export const maxDuration = 30
 
+// Data URL length cap (~3.75 MB of PNG), well above a normal diagram capture
+const MAX_IMAGE_DATA_LENGTH = 5 * 1024 * 1024
+
 interface ValidateDiagramRequest {
     imageData: string // Base64 PNG data URL
     sessionId?: string
@@ -25,25 +35,20 @@ const DEFAULT_VALID_RESULT: ValidationResult = {
     suggestions: [],
 }
 
-/**
- * Create a streaming response for useObject compatibility.
- * useObject expects text stream format, not plain JSON.
- */
+/** A fixed result in the text format useObject reads */
 function createStreamingResponse(result: ValidationResult): Response {
-    const encoder = new TextEncoder()
-    const stream = new ReadableStream({
-        start(controller) {
-            // Stream the JSON as text (useObject parses this)
-            controller.enqueue(encoder.encode(JSON.stringify(result)))
-            controller.close()
-        },
-    })
-    return new Response(stream, {
+    return new Response(JSON.stringify(result), {
         headers: { "Content-Type": "text/plain; charset=utf-8" },
     })
 }
 
 export async function POST(req: Request): Promise<Response> {
+    const crossSite = rejectCrossSite(req)
+    if (crossSite) return crossSite
+    // Uses the server's model credentials, so require the access code
+    const accessError = checkAccessCode(req)
+    if (accessError) return accessError
+
     try {
         // Check if VLM validation is enabled (default: true)
         const enableValidation = process.env.ENABLE_VLM_VALIDATION !== "false"
@@ -72,6 +77,42 @@ export async function POST(req: Request): Promise<Response> {
             )
         }
 
+        if (imageData.length > MAX_IMAGE_DATA_LENGTH) {
+            return Response.json(
+                { error: "Image data too large" },
+                { status: 413 },
+            )
+        }
+
+        // It runs the server's vision model: with the quota on, the daily
+        // and per-minute token limits apply, and its tokens are counted. Not
+        // the request limit, which is for chats: the day's last chat still
+        // gets its check, and a check does not count as a chat.
+        const userId = getUserIdFromRequest(req)
+        const countsQuota = isQuotaEnabled() && userId !== "anonymous"
+        if (countsQuota) {
+            const quotaCheck = await checkAndIncrementRequest(
+                userId,
+                {
+                    requests: 0,
+                    tokens: Number(process.env.DAILY_TOKEN_LIMIT) || 200000,
+                    tpm: Number(process.env.TPM_LIMIT) || 20000,
+                },
+                0,
+            )
+            if (!quotaCheck.allowed) {
+                return Response.json(
+                    {
+                        error: quotaCheck.error,
+                        type: quotaCheck.type,
+                        used: quotaCheck.used,
+                        limit: quotaCheck.limit,
+                    },
+                    { status: 429 },
+                )
+            }
+        }
+
         // Get the validation model
         let model
         try {
@@ -93,9 +134,9 @@ export async function POST(req: Request): Promise<Response> {
             ) || 10000
 
         // Stream the VLM response for useObject consumption
-        const result = streamObject({
+        const result = streamText({
             model,
-            schema: ValidationResultSchema,
+            output: Output.object({ schema: ValidationResultSchema }),
             system: VALIDATION_SYSTEM_PROMPT,
             messages: [
                 {
@@ -114,10 +155,17 @@ export async function POST(req: Request): Promise<Response> {
             ],
             maxOutputTokens: 1024,
             abortSignal: AbortSignal.timeout(timeout),
-            onFinish: ({ object }) => {
-                if (sessionId && object) {
+            onFinish: ({ output, totalUsage }) => {
+                if (countsQuota && totalUsage) {
+                    recordTokenUsage(
+                        userId,
+                        (totalUsage.inputTokens || 0) +
+                            (totalUsage.outputTokens || 0),
+                    )
+                }
+                if (sessionId && output) {
                     console.log(
-                        `[validate-diagram] Session ${sessionId}: valid=${object.valid}, issues=${object.issues?.length ?? 0}`,
+                        `[validate-diagram] Session ${sessionId}: valid=${output.valid}, issues=${output.issues?.length ?? 0}`,
                     )
                 }
             },

app/api/validate-model/route.ts

@@ -1,27 +1,28 @@
-import { createAmazonBedrock } from "@ai-sdk/amazon-bedrock"
-import { createAnthropic } from "@ai-sdk/anthropic"
-import { createDeepSeek, deepseek } from "@ai-sdk/deepseek"
-import { createGateway } from "@ai-sdk/gateway"
-import { createGoogleGenerativeAI } from "@ai-sdk/google"
-import { createVertex } from "@ai-sdk/google-vertex"
-import { createOpenAI } from "@ai-sdk/openai"
-import { createAihubmix } from "@aihubmix/ai-sdk-provider"
-import { createOpenRouter } from "@openrouter/ai-sdk-provider"
-import { generateText } from "ai"
+import { streamText, tool } from "ai"
 import { NextResponse } from "next/server"
-import { createOllama } from "ollama-ai-provider-v2"
+import { z } from "zod"
+import { checkAccessCode, rejectCrossSite } from "@/lib/access-code"
+import { checkAdminAuth } from "@/lib/admin/auth"
 import {
-    AIHUBMIX_APP_CODE,
-    isAihubmixStandardBaseURL,
-    normalizeMiniMaxBaseURL,
+    edgeOneEndpoint,
+    getAIModel,
+    globalBaseUrl,
+    usesServerCredentials,
+    usesServerEndpoint,
 } from "@/lib/ai-providers"
+import {
+    checkAndIncrementRequest,
+    isQuotaEnabled,
+} from "@/lib/dynamo-quota-manager"
+import { classifyLLMError } from "@/lib/llm-errors"
 import { allowPrivateUrls, isPrivateUrl } from "@/lib/ssrf-protection"
-import { PROVIDER_INFO, type ProviderName } from "@/lib/types/model-config"
+import { normalizeBaseUrl, type ProviderName } from "@/lib/types/model-config"
+import { getUserIdFromRequest } from "@/lib/user-id"
 
 export const runtime = "nodejs"
 
 interface ValidateRequest {
-    provider: string
+    provider: ProviderName
     apiKey: string
     baseUrl?: string
     modelId: string
@@ -29,21 +30,42 @@ interface ValidateRequest {
     awsAccessKeyId?: string
     awsSecretAccessKey?: string
     awsRegion?: string
+    awsSessionToken?: string
     // Vertex AI specific
     vertexApiKey?: string // Express Mode API key
+    // Set by the admin panel's Test: baseUrl is the server's <P>_BASE_URL
+    serverBaseUrl?: boolean
 }
 
+const TEST_TIMEOUT_MS = 15_000
+
+// Drawing works through tool calls, so the test asks for one
+const PING_TOOL = tool({
+    description: "Report that the connection works.",
+    inputSchema: z.object({}),
+})
+
+const NO_TOOL_CALL_WARNING =
+    "Connected, but the model answered without calling a tool. It may not support tool calls, which drawing needs."
+
 export async function POST(req: Request) {
+    const crossSite = rejectCrossSite(req)
+    if (crossSite) return crossSite
+    // Lets the server send requests to arbitrary URLs, so require the access
+    // code, or the admin password (the admin panel's Test button)
+    const accessError = checkAccessCode(req)
+    if (accessError && checkAdminAuth(req)) return accessError
+
     try {
         const body: ValidateRequest = await req.json()
         const {
             provider,
             apiKey,
-            baseUrl,
             modelId,
             awsAccessKeyId,
             awsSecretAccessKey,
             awsRegion,
+            awsSessionToken,
             // Note: Express Mode only needs vertexApiKey
             vertexApiKey,
         } = body
@@ -54,9 +76,26 @@ export async function POST(req: Request) {
                 { status: 400 },
             )
         }
+        // EdgeOne is this site's own function, as in the chat; the admin
+        // panel's Test sends no URL, and a relative one cannot be fetched
+        const baseUrl =
+            provider === "edgeone" ? edgeOneEndpoint(req) : body.baseUrl
+        // The admin panel's Test of an entry without a URL sends the
+        // server's own <P>_BASE_URL, which chat uses as it is: not a URL a
+        // user chose, so no private-address or redirect rules
+        const serverUrl =
+            body.serverBaseUrl === true &&
+            !!baseUrl &&
+            baseUrl === globalBaseUrl(provider) &&
+            !checkAdminAuth(req)
 
         // SECURITY: Block SSRF attacks via custom baseUrl
-        if (baseUrl && !allowPrivateUrls() && (await isPrivateUrl(baseUrl))) {
+        if (
+            baseUrl &&
+            !serverUrl &&
+            !allowPrivateUrls() &&
+            (await isPrivateUrl(baseUrl))
+        ) {
             return NextResponse.json(
                 { valid: false, error: "Invalid base URL" },
                 { status: 400 },
@@ -90,369 +129,117 @@ export async function POST(req: Request) {
                 { status: 400 },
             )
         }
-
-        let model: any
-
-        switch (provider) {
-            case "openai": {
-                const openai = createOpenAI({
-                    apiKey,
-                    ...(baseUrl && { baseURL: baseUrl }),
-                })
-                model = openai.chat(modelId)
-                break
-            }
-
-            case "anthropic": {
-                const anthropic = createAnthropic({
-                    apiKey,
-                    baseURL: baseUrl || "https://api.anthropic.com/v1",
-                })
-                model = anthropic(modelId)
-                break
-            }
-
-            case "google": {
-                const google = createGoogleGenerativeAI({
-                    apiKey,
-                    ...(baseUrl && { baseURL: baseUrl }),
-                })
-                model = google(modelId)
-                break
-            }
-
-            case "vertexai": {
-                const vertex = createVertex({
-                    apiKey: vertexApiKey,
-                    ...(baseUrl && { baseURL: baseUrl }),
-                })
-                model = vertex(modelId)
-                break
-            }
-
-            case "azure": {
-                const azure = createOpenAI({
-                    apiKey,
-                    baseURL: baseUrl,
-                })
-                model = azure.chat(modelId)
-                break
-            }
-
-            case "bedrock": {
-                const bedrock = createAmazonBedrock({
-                    accessKeyId: awsAccessKeyId,
-                    secretAccessKey: awsSecretAccessKey,
-                    region: awsRegion,
-                })
-                model = bedrock(modelId)
-                break
-            }
-
-            case "openrouter": {
-                const openrouter = createOpenRouter({
-                    apiKey,
-                    ...(baseUrl && { baseURL: baseUrl }),
-                })
-                model = openrouter(modelId)
-                break
-            }
-
-            case "aihubmix": {
-                const defaultBaseURL = PROVIDER_INFO.aihubmix.defaultBaseUrl
-
-                if (
-                    isAihubmixStandardBaseURL(baseUrl) ||
-                    baseUrl === defaultBaseURL
-                ) {
-                    const aihubmix = createAihubmix({
-                        apiKey,
-                        appCode: AIHUBMIX_APP_CODE,
-                    })
-                    model = aihubmix(modelId)
-                } else {
-                    const aihubmixCompatible = createOpenAI({
-                        apiKey,
-                        baseURL: baseUrl,
-                    })
-                    model = aihubmixCompatible.chat(modelId)
-                }
-                break
-            }
-
-            case "deepseek": {
-                if (baseUrl || apiKey) {
-                    const ds = createDeepSeek({
-                        apiKey,
-                        ...(baseUrl && { baseURL: baseUrl }),
-                    })
-                    model = ds(modelId)
-                } else {
-                    model = deepseek(modelId)
-                }
-                break
-            }
-
-            case "siliconflow": {
-                const sf = createOpenAI({
-                    apiKey,
-                    baseURL: baseUrl || "https://api.siliconflow.cn/v1",
-                })
-                model = sf.chat(modelId)
-                break
-            }
-
-            case "ollama": {
-                // SECURITY: Mirror ai-providers.ts guard — only use server
-                // OLLAMA_API_KEY when the URL is also from server config.
-                const ollamaApiKey = baseUrl
-                    ? apiKey || undefined
-                    : apiKey || process.env.OLLAMA_API_KEY || undefined
-                const ollamaProvider = createOllama({
-                    baseURL:
-                        baseUrl ||
-                        process.env.OLLAMA_BASE_URL ||
-                        "https://ollama.com/api",
-                    ...(ollamaApiKey && {
-                        headers: { Authorization: `Bearer ${ollamaApiKey}` },
-                    }),
-                })
-                model = ollamaProvider(modelId)
-                break
-            }
-
-            case "gateway": {
-                const gw = createGateway({
-                    apiKey,
-                    ...(baseUrl && { baseURL: baseUrl }),
-                })
-                model = gw(modelId)
-                break
-            }
-
-            case "edgeone": {
-                // EdgeOne uses OpenAI-compatible API via Edge Functions
-                // Need to pass cookies for EdgeOne Pages authentication
-                const cookieHeader = req.headers.get("cookie") || ""
-                const edgeone = createOpenAI({
-                    apiKey: "edgeone", // EdgeOne doesn't require API key
-                    baseURL: baseUrl || "/api/edgeai",
-                    headers: {
-                        cookie: cookieHeader,
-                    },
-                })
-                model = edgeone.chat(modelId)
-                break
-            }
-
-            case "sglang": {
-                // SGLang is OpenAI-compatible
-                const sglang = createOpenAI({
-                    apiKey: apiKey || "not-needed",
-                    baseURL: baseUrl || "http://127.0.0.1:8000/v1",
-                })
-                model = sglang.chat(modelId)
-                break
-            }
-
-            case "doubao": {
-                // ByteDance Doubao: use DeepSeek for DeepSeek/Kimi models, OpenAI for others
-                const doubaoBaseUrl =
-                    baseUrl || "https://ark.cn-beijing.volces.com/api/v3"
-                const lowerModelId = modelId.toLowerCase()
-                if (
-                    lowerModelId.includes("deepseek") ||
-                    lowerModelId.includes("kimi")
-                ) {
-                    const doubao = createDeepSeek({
-                        apiKey,
-                        baseURL: doubaoBaseUrl,
-                    })
-                    model = doubao(modelId)
-                } else {
-                    const doubao = createOpenAI({
-                        apiKey,
-                        baseURL: doubaoBaseUrl,
-                    })
-                    model = doubao.chat(modelId)
-                }
-                break
-            }
-
-            case "modelscope": {
-                const baseURL =
-                    baseUrl || "https://api-inference.modelscope.cn/v1"
-                const startTime = Date.now()
-
-                try {
-                    // Initiate a streaming request (required for QwQ-32B and certain Qwen3 models)
-                    const response = await fetch(
-                        `${baseURL}/chat/completions`,
-                        {
-                            method: "POST",
-                            headers: {
-                                "Content-Type": "application/json",
-                                Authorization: `Bearer ${apiKey}`,
-                            },
-                            body: JSON.stringify({
-                                model: modelId,
-                                messages: [
-                                    { role: "user", content: "Say 'OK'" },
-                                ],
-                                max_tokens: 20,
-                                stream: true,
-                                enable_thinking: false,
-                            }),
-                        },
-                    )
-
-                    if (!response.ok) {
-                        const errorText = await response.text()
-                        throw new Error(
-                            `ModelScope API error (${response.status}): ${errorText}`,
-                        )
-                    }
-
-                    const contentType =
-                        response.headers.get("content-type") || ""
-                    const isValidStreamingResponse =
-                        response.status === 200 &&
-                        (contentType.includes("text/event-stream") ||
-                            contentType.includes("application/json"))
-
-                    if (!isValidStreamingResponse) {
-                        throw new Error(
-                            `Unexpected response format: ${contentType}`,
-                        )
-                    }
-
-                    const responseTime = Date.now() - startTime
-
-                    if (response.body) {
-                        response.body.cancel().catch(() => {
-                            /* Ignore cancellation errors */
-                        })
-                    }
-
-                    return NextResponse.json({
-                        valid: true,
-                        responseTime,
-                        note: "ModelScope model validated (using streaming API)",
-                    })
-                } catch (error) {
-                    console.error(
-                        "[validate-model] ModelScope validation failed:",
-                        error,
-                    )
-                    throw error
-                }
-            }
-
-            case "minimax": {
-                const rawUrl =
-                    baseUrl ||
-                    PROVIDER_INFO.minimax?.defaultBaseUrl ||
-                    "https://api.minimaxi.com/anthropic"
-                const { baseURL: minimaxBaseUrl, isAnthropicCompatible } =
-                    normalizeMiniMaxBaseURL(rawUrl)
-
-                if (isAnthropicCompatible) {
-                    const minimax = createAnthropic({
-                        apiKey,
-                        baseURL: minimaxBaseUrl,
-                    })
-                    model = minimax.chat(modelId)
-                } else {
-                    const minimax = createOpenAI({
-                        apiKey,
-                        baseURL: minimaxBaseUrl,
-                    })
-                    model = minimax.chat(modelId)
-                }
-                break
-            }
-
-            // GLM, Qwen, Kimi, Qiniu, Novita, MiMo, Atlas Cloud - OpenAI compatible
-            case "glm":
-            case "qwen":
-            case "kimi":
-            case "qiniu":
-            case "novita":
-            case "atlascloud":
-            case "mimo": {
-                const baseURL =
-                    baseUrl ||
-                    PROVIDER_INFO[provider as ProviderName]?.defaultBaseUrl ||
-                    ""
-
-                if (!baseURL) {
-                    return NextResponse.json(
-                        {
-                            valid: false,
-                            error: `No base URL configured for provider: ${provider}`,
-                        },
-                        { status: 400 },
-                    )
-                }
-
-                const openai = createOpenAI({
-                    apiKey,
-                    baseURL,
-                })
-                model = openai.chat(modelId)
-                break
-            }
-
-            default:
-                return NextResponse.json(
-                    { valid: false, error: `Unknown provider: ${provider}` },
-                    { status: 400 },
-                )
+        // The Test button checks the user's own provider. On the server's
+        // keys (Ollama Cloud without a key or URL) anyone could run any model.
+        if (
+            usesServerCredentials(provider, {
+                apiKey,
+                baseUrl,
+                awsAccessKeyId,
+                awsSecretAccessKey,
+                vertexApiKey,
+            })
+        ) {
+            return NextResponse.json(
+                { valid: false, error: "API key is required" },
+                { status: 400 },
+            )
         }
 
-        // Make a minimal test request
-        const startTime = Date.now()
-        await generateText({
-            model,
-            prompt: "Say 'OK'",
-            maxOutputTokens: 20,
+        // On the deployment's own endpoints a Test runs a model as a chat
+        // does, so with the quota on it counts as a chat request (an
+        // admin's Test of the server's URL does not)
+        const userId = getUserIdFromRequest(req)
+        if (
+            isQuotaEnabled() &&
+            !serverUrl &&
+            userId !== "anonymous" &&
+            (await usesServerEndpoint(
+                provider,
+                normalizeBaseUrl(body.baseUrl ?? ""),
+                apiKey,
+            ))
+        ) {
+            const quotaCheck = await checkAndIncrementRequest(userId, {
+                requests: Number(process.env.DAILY_REQUEST_LIMIT) || 10,
+                tokens: Number(process.env.DAILY_TOKEN_LIMIT) || 200000,
+                tpm: Number(process.env.TPM_LIMIT) || 20000,
+            })
+            if (!quotaCheck.allowed) {
+                return NextResponse.json(
+                    { valid: false, error: quotaCheck.error },
+                    { status: 429 },
+                )
+            }
+        }
+
+        // The same model the chat would use. A client base URL makes it
+        // refuse redirects to internal hosts.
+        const { model } = getAIModel({
+            provider,
+            modelId,
+            apiKey,
+            baseUrl,
+            trustedBaseUrl: serverUrl,
+            awsAccessKeyId,
+            awsSecretAccessKey,
+            awsRegion,
+            // Temporary AWS credentials need it, as in the chat
+            awsSessionToken,
+            vertexApiKey,
+            // EdgeOne checks the Pages cookies and the access code
+            ...(provider === "edgeone" && {
+                headers: {
+                    cookie: req.headers.get("cookie") || "",
+                    "x-access-code": req.headers.get("x-access-code") || "",
+                },
+            }),
         })
+
+        // Streaming, like the chat (some models only stream). Stop at the
+        // first tool call; a reasoning model that runs out of tokens first
+        // proves the connection but not tool support.
+        const startTime = Date.now()
+        const result = streamText({
+            model,
+            prompt: "Call the ping tool.",
+            tools: { ping: PING_TOOL },
+            maxOutputTokens: 1024,
+            maxRetries: 0,
+            abortSignal: AbortSignal.timeout(TEST_TIMEOUT_MS),
+        })
+        let calledTool = false
+        let finishReason: string | undefined
+        for await (const part of result.fullStream) {
+            if (part.type === "error") throw part.error
+            // The timeout ends the stream with an abort part, not an error
+            if (part.type === "abort") {
+                const timeout = new Error(
+                    `The model did not answer within ${TEST_TIMEOUT_MS / 1000} s.`,
+                )
+                timeout.name = "TimeoutError"
+                throw timeout
+            }
+            if (part.type === "tool-call") {
+                calledTool = true
+                break
+            }
+            if (part.type === "finish") finishReason = part.finishReason
+        }
         const responseTime = Date.now() - startTime
 
         return NextResponse.json({
             valid: true,
             responseTime,
+            ...(!calledTool &&
+                finishReason !== "length" && { warning: NO_TOOL_CALL_WARNING }),
         })
     } catch (error) {
         console.error("[validate-model] Error:", error)
 
-        let errorMessage = "Validation failed"
-        if (error instanceof Error) {
-            // Extract meaningful error message
-            if (
-                error.message.includes("401") ||
-                error.message.includes("Unauthorized")
-            ) {
-                errorMessage = "Invalid API key"
-            } else if (
-                error.message.includes("404") ||
-                error.message.includes("not found")
-            ) {
-                errorMessage = "Model not found"
-            } else if (
-                error.message.includes("429") ||
-                error.message.includes("rate limit")
-            ) {
-                errorMessage = "Rate limited - try again later"
-            } else if (error.message.includes("ECONNREFUSED")) {
-                errorMessage = "Cannot connect to server"
-            } else {
-                errorMessage = error.message.slice(0, 100)
-            }
-        }
-
+        const { code, message } = classifyLLMError(error)
         return NextResponse.json(
-            { valid: false, error: errorMessage },
+            { valid: false, code, error: message },
             { status: 200 }, // Return 200 so client can read error message
         )
     }

app/api/verify-access-code/route.ts

@@ -1,29 +1,9 @@
+import { checkAccessCode } from "@/lib/access-code"
+
 export async function POST(req: Request) {
-    const accessCodes =
-        process.env.ACCESS_CODE_LIST?.split(",")
-            .map((code) => code.trim())
-            .filter(Boolean) || []
-
-    // If no access codes configured, verification always passes
-    if (accessCodes.length === 0) {
-        return Response.json({
-            valid: true,
-            message: "No access code required",
-        })
-    }
-
-    const accessCodeHeader = req.headers.get("x-access-code")
-
-    if (!accessCodeHeader) {
+    if (checkAccessCode(req)) {
         return Response.json(
-            { valid: false, message: "Access code is required" },
-            { status: 401 },
-        )
-    }
-
-    if (!accessCodes.includes(accessCodeHeader)) {
-        return Response.json(
-            { valid: false, message: "Invalid access code" },
+            { valid: false, message: "Invalid or missing access code" },
             { status: 401 },
         )
     }

biome.json

@@ -7,7 +7,12 @@
     },
     "files": {
         "ignoreUnknown": false,
-        "includes": ["**", "!public"]
+        "includes": [
+            "**",
+            "!public",
+            "!packages/mcp-server/src/preview",
+            "!lib/model-catalog.json"
+        ]
     },
     "formatter": {
         "enabled": true,

components/chat-example-panel.tsx

@@ -77,6 +77,7 @@ export default function ExamplePanel({
     minimal?: boolean
 }) {
     const dict = useDictionary()
+    const isSelfHosted = process.env.NEXT_PUBLIC_SELFHOSTED === "true"
 
     const handleReplicateFlowchart = async () => {
         setInput("Replicate this flowchart.")
@@ -125,29 +126,31 @@ export default function ExamplePanel({
         <div className={minimal ? "" : "py-6 px-2 animate-fade-in"}>
             {!minimal && (
                 <>
-                    {/* MCP Server Notice */}
-                    <a
-                        href="https://github.com/DayuanJiang/next-ai-draw-io/tree/main/packages/mcp-server"
-                        target="_blank"
-                        rel="noopener noreferrer"
-                        className="block mb-4 p-3 rounded-xl bg-gradient-to-r from-purple-500/10 to-blue-500/10 border border-purple-500/20 hover:border-purple-500/40 transition-colors group"
-                    >
-                        <div className="flex items-center gap-3">
-                            <div className="w-8 h-8 rounded-lg bg-purple-500/20 flex items-center justify-center shrink-0">
-                                <Terminal className="w-4 h-4 text-purple-500" />
-                            </div>
-                            <div className="min-w-0">
-                                <div className="flex items-center gap-2">
-                                    <span className="text-sm font-medium text-foreground group-hover:text-purple-500 transition-colors">
-                                        {dict.examples.mcpServer}
-                                    </span>
+                    {/* MCP Server Notice, hidden on self-hosted deployments */}
+                    {!isSelfHosted && (
+                        <a
+                            href="https://github.com/DayuanJiang/next-ai-draw-io/tree/main/packages/mcp-server"
+                            target="_blank"
+                            rel="noopener noreferrer"
+                            className="block mb-4 p-3 rounded-xl bg-gradient-to-r from-purple-500/10 to-blue-500/10 border border-purple-500/20 hover:border-purple-500/40 transition-colors group"
+                        >
+                            <div className="flex items-center gap-3">
+                                <div className="w-8 h-8 rounded-lg bg-purple-500/20 flex items-center justify-center shrink-0">
+                                    <Terminal className="w-4 h-4 text-purple-500" />
+                                </div>
+                                <div className="min-w-0">
+                                    <div className="flex items-center gap-2">
+                                        <span className="text-sm font-medium text-foreground group-hover:text-purple-500 transition-colors">
+                                            {dict.examples.mcpServer}
+                                        </span>
+                                    </div>
+                                    <p className="text-xs text-muted-foreground">
+                                        {dict.examples.mcpDescription}
+                                    </p>
                                 </div>
-                                <p className="text-xs text-muted-foreground">
-                                    {dict.examples.mcpDescription}
-                                </p>
                             </div>
-                        </div>
-                    </a>
+                        </a>
+                    )}
 
                     {/* Welcome section */}
                     <div className="text-center mb-6">

components/chat-input.tsx

@@ -11,7 +11,9 @@ import {
 } from "lucide-react"
 import type React from "react"
 import {
+    type Dispatch,
     forwardRef,
+    type SetStateAction,
     useCallback,
     useEffect,
     useImperativeHandle,
@@ -41,9 +43,20 @@ import { FilePreviewList } from "./file-preview-list"
 
 const MAX_IMAGE_SIZE = 2 * 1024 * 1024 // 2MB
 const MAX_FILES = 5
+// Image formats every supported model provider accepts (SVG is read as text)
+const SUPPORTED_IMAGE_TYPES = [
+    "image/png",
+    "image/jpeg",
+    "image/gif",
+    "image/webp",
+]
 
 function isValidFileType(file: File): boolean {
-    return file.type.startsWith("image/") || isPdfFile(file) || isTextFile(file)
+    return (
+        SUPPORTED_IMAGE_TYPES.includes(file.type) ||
+        isPdfFile(file) ||
+        isTextFile(file)
+    )
 }
 
 function formatFileSize(bytes: number): string {
@@ -164,7 +177,7 @@ interface ChatInputProps {
         { text: string; charCount: number; isExtracting: boolean }
     >
     urlData?: Map<string, UrlData>
-    onUrlChange?: (data: Map<string, UrlData>) => void
+    onUrlChange?: Dispatch<SetStateAction<Map<string, UrlData>>>
 
     sessionId?: string
     error?: Error | null
@@ -244,6 +257,11 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
         // Allow retry when there's an error (even if status is still "streaming" or "submitted")
         const isDisabled =
             (status === "streaming" || status === "submitted") && !error
+        // Block sending until attached files and URLs have their text, otherwise
+        // their content would be silently dropped
+        const isExtractingAttachments =
+            files.some((file) => pdfData.get(file)?.isExtracting) ||
+            Array.from(urlData?.values() ?? []).some((d) => d.isExtracting)
 
         const adjustTextareaHeight = useCallback(() => {
             const textarea = textareaRef.current
@@ -281,6 +299,9 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
         }
 
         const handleKeyDown = (e: React.KeyboardEvent) => {
+            // Enter that confirms an IME candidate must not send the message
+            if (e.nativeEvent.isComposing || e.keyCode === 229) return
+
             const shouldSend =
                 sendShortcut === "enter"
                     ? e.key === "Enter" &&
@@ -292,7 +313,12 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
             if (shouldSend) {
                 e.preventDefault()
                 const form = e.currentTarget.closest("form")
-                if (form && input.trim() && !isDisabled) {
+                if (
+                    form &&
+                    input.trim() &&
+                    !isDisabled &&
+                    !isExtractingAttachments
+                ) {
                     form.requestSubmit()
                 }
             }
@@ -380,13 +406,9 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
 
             if (isDisabled) return
 
-            const droppedFiles = e.dataTransfer.files
-            const supportedFiles = Array.from(droppedFiles).filter((file) =>
-                isValidFileType(file),
-            )
-
+            // Let validateFiles show a toast for unsupported types
             const { validFiles, errors } = validateFiles(
-                supportedFiles,
+                Array.from(e.dataTransfer.files),
                 files.length,
                 dict,
             )
@@ -401,33 +423,34 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
 
             setIsExtractingUrl(true)
 
+            // Use functional updates so a removal or send made while extracting
+            // is not overwritten when the request finishes
             try {
-                const existing = urlData
-                    ? new Map(urlData)
-                    : new Map<string, UrlData>()
-                existing.set(url, {
-                    url,
-                    title: url,
-                    content: "",
-                    charCount: 0,
-                    isExtracting: true,
-                })
-                onUrlChange(existing)
+                onUrlChange((prev) =>
+                    new Map(prev).set(url, {
+                        url,
+                        title: url,
+                        content: "",
+                        charCount: 0,
+                        isExtracting: true,
+                    }),
+                )
 
                 const data = await extractUrlContent(url)
 
-                const newUrlData = new Map(existing)
-                newUrlData.set(url, data)
-                onUrlChange(newUrlData)
+                // Skip if the URL was removed while extracting
+                onUrlChange((prev) =>
+                    prev.has(url) ? new Map(prev).set(url, data) : prev,
+                )
 
                 setShowUrlDialog(false)
             } catch (error) {
                 // Remove the URL from the data map on error
-                const newUrlData = urlData
-                    ? new Map(urlData)
-                    : new Map<string, UrlData>()
-                newUrlData.delete(url)
-                onUrlChange(newUrlData)
+                onUrlChange((prev) => {
+                    const next = new Map(prev)
+                    next.delete(url)
+                    return next
+                })
                 showErrorToast(
                     <span className="text-muted-foreground">
                         {error instanceof Error
@@ -463,11 +486,12 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
                             urlData={urlData}
                             onRemoveUrl={
                                 onUrlChange
-                                    ? (url) => {
-                                          const next = new Map(urlData)
-                                          next.delete(url)
-                                          onUrlChange(next)
-                                      }
+                                    ? (url) =>
+                                          onUrlChange((prev) => {
+                                              const next = new Map(prev)
+                                              next.delete(url)
+                                              return next
+                                          })
                                     : undefined
                             }
                         />
@@ -559,7 +583,7 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
                                 ref={fileInputRef}
                                 className="hidden"
                                 onChange={handleFileChange}
-                                accept="image/*,.pdf,application/pdf,text/*,.md,.markdown,.json,.csv,.xml,.yaml,.yml,.toml"
+                                accept="image/png,image/jpeg,image/gif,image/webp,.svg,.pdf,application/pdf,text/*,.md,.markdown,.json,.csv,.xml,.yaml,.yml,.toml"
                                 multiple
                                 disabled={isDisabled}
                             />
@@ -588,7 +612,11 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
                         ) : (
                             <Button
                                 type="submit"
-                                disabled={isDisabled || !input.trim()}
+                                disabled={
+                                    isDisabled ||
+                                    isExtractingAttachments ||
+                                    !input.trim()
+                                }
                                 size="sm"
                                 className="h-8 px-4 rounded-xl font-medium shadow-sm"
                                 aria-label={dict.chat.send}
@@ -614,9 +642,8 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
                             dict.save.savedSuccessfully,
                         )
                     }
-                    defaultFilename={`diagram-${new Date()
-                        .toISOString()
-                        .slice(0, 10)}`}
+                    // Local date as YYYY-MM-DD (toISOString would give UTC)
+                    defaultFilename={`diagram-${new Date().toLocaleDateString("sv-SE")}`}
                 />
                 {onUrlChange && (
                     <UrlInputDialog
@@ -629,7 +656,11 @@ export const ChatInput = forwardRef<ChatInputRef, ChatInputProps>(
                 <TemplateCreateDialog
                     open={showSaveAsTemplate}
                     onOpenChange={setShowSaveAsTemplate}
-                    onSuccess={() => setShowSaveAsTemplate(false)}
+                    onSuccess={() => {
+                        setShowSaveAsTemplate(false)
+                        // Let the template list in the lobby reload
+                        window.dispatchEvent(new Event("templatesChanged"))
+                    }}
                     initialPrompt={input.trim()}
                 />
             </form>

components/chat-message-display.tsx

@@ -26,6 +26,7 @@ import {
     ReasoningContent,
     ReasoningTrigger,
 } from "@/components/ai-elements/reasoning"
+import { Shimmer } from "@/components/ai-elements/shimmer"
 import { ChatLobby } from "@/components/chat/ChatLobby"
 import { TemplateCreateDialog } from "@/components/chat/TemplateCreateDialog"
 import { ToolCallCard } from "@/components/chat/ToolCallCard"
@@ -37,12 +38,12 @@ import { ScrollArea } from "@/components/ui/scroll-area"
 import { useDictionary } from "@/hooks/use-dictionary"
 import { getApiEndpoint } from "@/lib/base-path"
 import {
-    applyDiagramOperations,
     convertToLegalXml,
     extractCompleteMxCells,
     replaceNodes,
-    validateAndFixXml,
 } from "@/lib/utils"
+import { applyDiagramOperations } from "@/packages/mcp-server/src/diagram-operations.ts"
+import { BLANK_MXFILE } from "@/packages/mcp-server/src/pages.ts"
 
 // Helper to extract complete operations from streaming input
 function getCompleteOperations(
@@ -129,12 +130,14 @@ const getMessageTextContent = (message: UIMessage): string => {
         .join("\n")
 }
 
+// Matches the [PDF: ...], [File: ...] and [URL: ...] sections appended to the user's text
+export const APPENDED_FILE_SECTIONS_PATTERN =
+    /\n\n\[(PDF|File|URL):\s*[^\]]+\]\n[\s\S]*$/
+
 // Get only the user's original text, excluding appended file content
 const getUserOriginalText = (message: UIMessage): string => {
     const fullText = getMessageTextContent(message)
-    // Strip out [PDF: ...], [File: ...], and [URL: ...] sections that were appended
-    const filePattern = /\n\n\[(PDF|File|URL):\s*[^\]]+\]\n[\s\S]*$/
-    return fullText.replace(filePattern, "").trim()
+    return fullText.replace(APPENDED_FILE_SECTIONS_PATTERN, "").trim()
 }
 
 interface SessionMetadata {
@@ -146,6 +149,8 @@ interface SessionMetadata {
 
 interface ChatMessageDisplayProps {
     messages: UIMessage[]
+    // Shown on an error that a model setting can fix (bad key, unknown model)
+    onOpenModelConfig?: () => void
     setInput: (input: string) => void
     setFiles: (files: File[]) => void
     processedToolCallsRef: MutableRefObject<Set<string>>
@@ -169,6 +174,7 @@ interface ChatMessageDisplayProps {
 
 export function ChatMessageDisplay({
     messages,
+    onOpenModelConfig,
     setInput,
     setFiles,
     processedToolCallsRef,
@@ -188,7 +194,24 @@ export function ChatMessageDisplay({
     currentInput = "",
 }: ChatMessageDisplayProps) {
     const dict = useDictionary()
-    const { chartXML, loadDiagram: onDisplayChart } = useDiagram()
+    // The thinking header in the page language
+    const thinkingMessage = (isStreaming: boolean, duration?: number) => {
+        if (isStreaming || duration === 0) {
+            return <Shimmer duration={1}>{dict.reasoning.thinking}</Shimmer>
+        }
+        if (duration === undefined) return <p>{dict.reasoning.thoughtBrief}</p>
+        return (
+            <p>
+                {duration === 1
+                    ? dict.reasoning.thoughtForOne
+                    : dict.reasoning.thoughtFor.replace(
+                          "{duration}",
+                          String(duration),
+                      )}
+            </p>
+        )
+    }
+    const { chartXML, chartXMLRef, loadDiagram: onDisplayChart } = useDiagram()
     const messagesEndRef = useRef<HTMLDivElement>(null)
     const scrollTopRef = useRef<HTMLDivElement>(null)
     const previousXML = useRef<string>("")
@@ -208,20 +231,6 @@ export function ChatMessageDisplay({
             scrollTopRef.current?.scrollIntoView({ behavior: "instant" })
         }
     }, [messages.length, processedToolCalls])
-    // Debounce streaming diagram updates - store pending XML and timeout
-    const pendingXmlRef = useRef<string | null>(null)
-    const debounceTimeoutRef = useRef<ReturnType<typeof setTimeout> | null>(
-        null,
-    )
-    const STREAMING_DEBOUNCE_MS = 150 // Only update diagram every 150ms during streaming
-    // Refs for edit_diagram streaming
-    const pendingEditRef = useRef<{
-        operations: DiagramOperation[]
-        toolCallId: string
-    } | null>(null)
-    const editDebounceTimeoutRef = useRef<ReturnType<typeof setTimeout> | null>(
-        null,
-    )
     const [expandedTools, setExpandedTools] = useState<Record<string, boolean>>(
         {},
     )
@@ -343,73 +352,30 @@ export function ChatMessageDisplay({
         }
     }
 
+    // Streaming preview of display_diagram: draw the complete cells written
+    // so far. The tool handler validates and loads the final diagram.
     const handleDisplayChart = useCallback(
-        (xml: string, showToast = false) => {
-            let currentXml = xml || ""
+        (xml: string) => {
+            const completeCells = extractCompleteMxCells(xml || "")
+            if (!completeCells) return
+            const convertedXml = convertToLegalXml(completeCells)
+            if (convertedXml === previousXML.current) return
 
-            // During streaming (showToast=false), extract only complete mxCell elements
-            // This allows progressive rendering even with partial/incomplete trailing XML
-            if (!showToast) {
-                const completeCells = extractCompleteMxCells(currentXml)
-                if (!completeCells) {
-                    return
-                }
-                currentXml = completeCells
-            }
+            // Skip this update while the cells written so far don't parse
+            const testDoc = new DOMParser().parseFromString(
+                `<root>${convertedXml}</root>`,
+                "text/xml",
+            )
+            if (testDoc.querySelector("parsererror")) return
 
-            const convertedXml = convertToLegalXml(currentXml)
-            if (convertedXml !== previousXML.current) {
-                // Parse and validate XML BEFORE calling replaceNodes
-                const parser = new DOMParser()
-                // Wrap in root element for parsing multiple mxCell elements
-                const testDoc = parser.parseFromString(
-                    `<root>${convertedXml}</root>`,
-                    "text/xml",
-                )
-                const parseError = testDoc.querySelector("parsererror")
-
-                if (parseError) {
-                    // Only show toast if this is the final XML (not during streaming)
-                    if (showToast) {
-                        toast.error(dict.errors.malformedXml)
-                    }
-                    return // Skip this update
-                }
-
-                try {
-                    // If chartXML is empty, create a default mxfile structure to use with replaceNodes
-                    // This ensures the XML is properly wrapped in mxfile/diagram/mxGraphModel format
-                    const baseXML =
-                        chartXML ||
-                        `<mxfile><diagram name="Page-1" id="page-1"><mxGraphModel><root><mxCell id="0"/><mxCell id="1" parent="0"/></root></mxGraphModel></diagram></mxfile>`
-                    const replacedXML = replaceNodes(baseXML, convertedXml)
-
-                    // During streaming (showToast=false), skip heavy validation for lower latency
-                    // The quick DOM parse check above catches malformed XML
-                    // Full validation runs on final output (showToast=true)
-                    if (!showToast) {
-                        previousXML.current = convertedXml
-                        onDisplayChart(replacedXML, true)
-                        return
-                    }
-
-                    // Final output: run full validation and auto-fix
-                    const validation = validateAndFixXml(replacedXML)
-                    if (validation.valid) {
-                        previousXML.current = convertedXml
-                        // Use fixed XML if available, otherwise use original
-                        const xmlToLoad = validation.fixed || replacedXML
-                        onDisplayChart(xmlToLoad, true)
-                    } else {
-                        toast.error(dict.errors.validationFailed)
-                    }
-                } catch (error) {
-                    console.error("Error processing XML:", error)
-                    // Only show toast if this is the final XML (not during streaming)
-                    if (showToast) {
-                        toast.error(dict.errors.failedToProcess)
-                    }
-                }
+            try {
+                // An empty canvas gets a default mxfile to put the cells in
+                const baseXML = chartXML || BLANK_MXFILE
+                const replacedXML = replaceNodes(baseXML, convertedXml)
+                previousXML.current = convertedXml
+                onDisplayChart(replacedXML, true)
+            } catch (error) {
+                console.error("Error processing XML:", error)
             }
         },
         [chartXML, onDisplayChart],
@@ -456,8 +422,19 @@ export function ChatMessageDisplay({
         // Previous messages are already processed and won't change
         const messagesToProcess =
             messages.length > 0 ? [messages[messages.length - 1]] : []
+        // The diagram without streamed previews, as loaded last: the tool
+        // handler's result of an earlier edit is there before the chartXML
+        // state catches up. Undoing a failed edit's preview below changes it
+        // too, and an edit streaming right after must start from the undone
+        // diagram.
+        let baseXml = chartXMLRef.current
 
         messagesToProcess.forEach((message) => {
+            // Messages restored from a saved session were applied before it was
+            // saved; the saved diagram is authoritative, so don't replay them
+            const isRestoredMessage =
+                loadedMessageIdsRef?.current.has(message.id) ?? false
+
             if (message.parts) {
                 message.parts.forEach((part) => {
                     if (part.type?.startsWith("tool-")) {
@@ -475,76 +452,87 @@ export function ChatMessageDisplay({
                             })
                         }
 
+                        if (isRestoredMessage) return
+
                         if (
-                            part.type === "tool-display_diagram" &&
-                            input?.xml
+                            part.type !== "tool-display_diagram" &&
+                            part.type !== "tool-edit_diagram"
                         ) {
-                            const xml = input.xml as string
+                            return
+                        }
 
+                        // Failed or stopped: if the original XML is still
+                        // stored, the tool handler never ran (invalid JSON,
+                        // or the user pressed stop), so undo the streamed
+                        // preview here. Invalid JSON leaves no input, so
+                        // check this first.
+                        if (state === "output-error") {
+                            const originalXml =
+                                editDiagramOriginalXmlRef.current.get(
+                                    toolCallId,
+                                )
+                            if (originalXml) {
+                                editDiagramOriginalXmlRef.current.delete(
+                                    toolCallId,
+                                )
+                                onDisplayChart(originalXml, true)
+                                baseXml = originalXml
+                            }
+                            return
+                        }
+
+                        // Input complete, or the tool handler, a stop or an
+                        // error took the call already: the tool handler loads
+                        // the checked diagram (with the original XML). The
+                        // messages update at most every 150 ms (useChat
+                        // throttle in chat-panel), so they can still show the
+                        // call streaming after that.
+                        if (
+                            state !== "input-streaming" ||
+                            processedToolCalls.current.has(toolCallId)
+                        ) {
+                            processedToolCalls.current.add(toolCallId)
+                            lastProcessedXmlRef.current.delete(toolCallId)
+                            lastProcessedXmlRef.current.delete(
+                                `${toolCallId}-opCount`,
+                            )
+                            return
+                        }
+
+                        if (part.type === "tool-display_diagram") {
+                            const xml = input?.xml as string | undefined
                             // Skip if XML hasn't changed since last processing
-                            const lastXml =
-                                lastProcessedXmlRef.current.get(toolCallId)
-                            if (lastXml === xml) {
-                                return // Skip redundant processing
-                            }
-
                             if (
-                                state === "input-streaming" ||
-                                state === "input-available"
+                                !xml ||
+                                lastProcessedXmlRef.current.get(toolCallId) ===
+                                    xml
                             ) {
-                                // Debounce streaming updates - queue the XML and process after delay
-                                pendingXmlRef.current = xml
-
-                                if (!debounceTimeoutRef.current) {
-                                    // No pending timeout - set one up
-                                    debounceTimeoutRef.current = setTimeout(
-                                        () => {
-                                            const pendingXml =
-                                                pendingXmlRef.current
-                                            debounceTimeoutRef.current = null
-                                            pendingXmlRef.current = null
-                                            if (pendingXml) {
-                                                handleDisplayChart(
-                                                    pendingXml,
-                                                    false,
-                                                )
-                                                lastProcessedXmlRef.current.set(
-                                                    toolCallId,
-                                                    pendingXml,
-                                                )
-                                            }
-                                        },
-                                        STREAMING_DEBOUNCE_MS,
-                                    )
-                                }
-                            } else if (
-                                state === "output-available" &&
-                                !processedToolCalls.current.has(toolCallId)
-                            ) {
-                                // Final output - process immediately (clear any pending debounce)
-                                if (debounceTimeoutRef.current) {
-                                    clearTimeout(debounceTimeoutRef.current)
-                                    debounceTimeoutRef.current = null
-                                    pendingXmlRef.current = null
-                                }
-                                // Show toast only if final XML is malformed
-                                handleDisplayChart(xml, true)
-                                processedToolCalls.current.add(toolCallId)
-                                // Clean up the ref entry - tool is complete, no longer needed
-                                lastProcessedXmlRef.current.delete(toolCallId)
+                                return
                             }
+                            // Keep the diagram from before the preview, to
+                            // undo it on a stop or an error
+                            if (
+                                !editDiagramOriginalXmlRef.current.has(
+                                    toolCallId,
+                                )
+                            ) {
+                                editDiagramOriginalXmlRef.current.set(
+                                    toolCallId,
+                                    baseXml || BLANK_MXFILE,
+                                )
+                            }
+                            handleDisplayChart(xml)
+                            lastProcessedXmlRef.current.set(toolCallId, xml)
+                            return
                         }
 
                         // Handle edit_diagram streaming - apply operations incrementally for preview
                         // Uses shared editDiagramOriginalXmlRef to coordinate with tool handler
-                        if (
-                            part.type === "tool-edit_diagram" &&
-                            input?.operations
-                        ) {
+                        if (part.type === "tool-edit_diagram") {
+                            if (!input?.operations) return
                             const completeOps = getCompleteOperations(
                                 input.operations as DiagramOperation[],
                             )
-
                             if (completeOps.length === 0) return
 
                             // Capture original XML when streaming starts (store in shared ref)
@@ -553,7 +541,7 @@ export function ChatMessageDisplay({
                                     toolCallId,
                                 )
                             ) {
-                                if (!chartXML) {
+                                if (!baseXml) {
                                     console.warn(
                                         "[edit_diagram streaming] No chart XML available",
                                     )
@@ -561,10 +549,9 @@ export function ChatMessageDisplay({
                                 }
                                 editDiagramOriginalXmlRef.current.set(
                                     toolCallId,
-                                    chartXML,
+                                    baseXml,
                                 )
                             }
-
                             const originalXml =
                                 editDiagramOriginalXmlRef.current.get(
                                     toolCallId,
@@ -572,95 +559,37 @@ export function ChatMessageDisplay({
                             if (!originalXml) return
 
                             // Skip if no change from last processed state
-                            const lastCount = lastProcessedXmlRef.current.get(
-                                toolCallId + "-opCount",
-                            )
-                            if (lastCount === String(completeOps.length)) return
-
+                            const countKey = `${toolCallId}-opCount`
+                            const opCount = String(completeOps.length)
                             if (
-                                state === "input-streaming" ||
-                                state === "input-available"
+                                lastProcessedXmlRef.current.get(countKey) ===
+                                opCount
                             ) {
-                                // Queue the operations for debounced processing
-                                pendingEditRef.current = {
-                                    operations: completeOps,
-                                    toolCallId,
-                                }
-
-                                if (!editDebounceTimeoutRef.current) {
-                                    editDebounceTimeoutRef.current = setTimeout(
-                                        () => {
-                                            const pending =
-                                                pendingEditRef.current
-                                            editDebounceTimeoutRef.current =
-                                                null
-                                            pendingEditRef.current = null
-
-                                            if (pending) {
-                                                const origXml =
-                                                    editDiagramOriginalXmlRef.current.get(
-                                                        pending.toolCallId,
-                                                    )
-                                                if (!origXml) return
-
-                                                try {
-                                                    const {
-                                                        result: editedXml,
-                                                    } = applyDiagramOperations(
-                                                        origXml,
-                                                        pending.operations,
-                                                    )
-                                                    handleDisplayChart(
-                                                        editedXml,
-                                                        false,
-                                                    )
-                                                    lastProcessedXmlRef.current.set(
-                                                        pending.toolCallId +
-                                                            "-opCount",
-                                                        String(
-                                                            pending.operations
-                                                                .length,
-                                                        ),
-                                                    )
-                                                } catch (e) {
-                                                    console.warn(
-                                                        `[edit_diagram streaming] Operation failed:`,
-                                                        e instanceof Error
-                                                            ? e.message
-                                                            : e,
-                                                    )
-                                                }
-                                            }
-                                        },
-                                        STREAMING_DEBOUNCE_MS,
-                                    )
-                                }
-                            } else if (
-                                state === "output-available" &&
-                                !processedToolCalls.current.has(toolCallId)
-                            ) {
-                                // Final state - cleanup streaming refs (tool handler does final application)
-                                if (editDebounceTimeoutRef.current) {
-                                    clearTimeout(editDebounceTimeoutRef.current)
-                                    editDebounceTimeoutRef.current = null
-                                }
-                                lastProcessedXmlRef.current.delete(
-                                    toolCallId + "-opCount",
+                                return
+                            }
+                            try {
+                                const { result } = applyDiagramOperations(
+                                    originalXml,
+                                    completeOps,
+                                )
+                                // Load the full document so other pages stay intact
+                                onDisplayChart(result, true)
+                                lastProcessedXmlRef.current.set(
+                                    countKey,
+                                    opCount,
+                                )
+                            } catch (e) {
+                                console.warn(
+                                    "[edit_diagram streaming] Operation failed:",
+                                    e instanceof Error ? e.message : e,
                                 )
-                                processedToolCalls.current.add(toolCallId)
-                                // Note: Don't delete editDiagramOriginalXmlRef here - tool handler needs it
                             }
                         }
                     }
                 })
             }
         })
-
-        // NOTE: Don't cleanup debounce timeouts here!
-        // The cleanup runs on every re-render (when messages changes),
-        // which would cancel the timeout before it fires.
-        // Let the timeouts complete naturally - they're harmless if component unmounts.
-    }, [messages, handleDisplayChart, chartXML])
+    }, [messages, handleDisplayChart, chartXMLRef])
 
     return (
         <ScrollArea className="h-full w-full scrollbar-thin">
@@ -845,7 +774,11 @@ export function ChatMessageDisplay({
                                                                 !isRestoredMessage
                                                             }
                                                         >
-                                                            <ReasoningTrigger />
+                                                            <ReasoningTrigger
+                                                                getThinkingMessage={
+                                                                    thinkingMessage
+                                                                }
+                                                            />
                                                             <ReasoningContent>
                                                                 {
                                                                     reasoningPart.text
@@ -951,8 +884,12 @@ export function ChatMessageDisplay({
                                                     part.type?.startsWith(
                                                         "tool-",
                                                     )
+                                                // Blank text (some models send
+                                                // a lone space) gets no bubble
                                                 const isContentPart =
-                                                    part.type === "text" ||
+                                                    (part.type === "text" &&
+                                                        part.text.trim() !==
+                                                            "") ||
                                                     part.type === "file"
 
                                                 if (isToolPart) {
@@ -1302,6 +1239,32 @@ export function ChatMessageDisplay({
                                                                     return null
                                                                 },
                                                             )}
+                                                            {message.role ===
+                                                                "system" &&
+                                                                (
+                                                                    message.metadata as
+                                                                        | {
+                                                                              openModelConfig?: boolean
+                                                                          }
+                                                                        | undefined
+                                                                )
+                                                                    ?.openModelConfig &&
+                                                                onOpenModelConfig && (
+                                                                    <button
+                                                                        type="button"
+                                                                        onClick={
+                                                                            onOpenModelConfig
+                                                                        }
+                                                                        className="mt-2 text-xs font-medium underline underline-offset-2 hover:opacity-80"
+                                                                    >
+                                                                        {
+                                                                            dict
+                                                                                .errors
+                                                                                .llm
+                                                                                .openModelSettings
+                                                                        }
+                                                                    </button>
+                                                                )}
                                                         </div>
                                                     )
                                                 },

components/chat-panel.tsx

@@ -1,7 +1,7 @@
 "use client"
 
 import { useChat } from "@ai-sdk/react"
-import { DefaultChatTransport } from "ai"
+import { DefaultChatTransport, isToolUIPart, type UIMessage } from "ai"
 import {
     MessageSquarePlus,
     PanelRightClose,
@@ -41,8 +41,13 @@ import type { UrlData } from "@/lib/url-utils"
 import { type FileData, useFileProcessor } from "@/lib/use-file-processor"
 import { useQuotaManager } from "@/lib/use-quota-manager"
 import { cn, formatXML, isRealDiagram } from "@/lib/utils"
+import { prepareNewDiagram } from "@/packages/mcp-server/src/new-diagram.ts"
+import { BLANK_MXFILE, hasCells } from "@/packages/mcp-server/src/pages.ts"
 import type { ValidationState } from "./chat/ValidationCard"
-import { ChatMessageDisplay } from "./chat-message-display"
+import {
+    APPENDED_FILE_SECTIONS_PATTERN,
+    ChatMessageDisplay,
+} from "./chat-message-display"
 import { DevXmlSimulator } from "./dev-xml-simulator"
 
 // localStorage keys for persistence
@@ -51,21 +56,6 @@ const STORAGE_SESSION_ID_KEY = "next-ai-draw-io-session-id"
 // sessionStorage keys
 const SESSION_STORAGE_INPUT_KEY = "next-ai-draw-io-input"
 
-// Type for message parts (tool calls and their states)
-interface MessagePart {
-    type: string
-    state?: string
-    toolName?: string
-    input?: { xml?: string; [key: string]: unknown }
-    [key: string]: unknown
-}
-
-interface ChatMessage {
-    role: string
-    parts?: MessagePart[]
-    [key: string]: unknown
-}
-
 interface ChatPanelProps {
     isVisible: boolean
     onToggleVisibility: () => void
@@ -88,25 +78,25 @@ const MAX_CONTINUATION_RETRY_COUNT = 2 // Limit for truncation continuation retr
  * Check if auto-resubmit should happen based on tool errors.
  * Only checks the LAST tool part (most recent tool call), not all tool parts.
  */
-function hasToolErrors(messages: ChatMessage[]): boolean {
+function hasToolErrors(messages: UIMessage[]): boolean {
     const lastMessage = messages[messages.length - 1]
-    if (!lastMessage || lastMessage.role !== "assistant") {
-        return false
-    }
-
-    const toolParts =
-        (lastMessage.parts as MessagePart[] | undefined)?.filter((part) =>
-            part.type?.startsWith("tool-"),
-        ) || []
-
-    if (toolParts.length === 0) {
-        return false
-    }
-
-    const lastToolPart = toolParts[toolParts.length - 1]
+    if (lastMessage?.role !== "assistant") return false
+    const lastToolPart = lastMessage.parts.filter(isToolUIPart).at(-1)
     return lastToolPart?.state === TOOL_ERROR_STATE
 }
 
+/**
+ * Snapshots keep the full multi-page document, but the model only sees and
+ * edits the first page, so give it the first page's mxGraphModel.
+ * Older snapshots already hold a single mxGraphModel and are returned as is.
+ */
+function getFirstPageXml(xml: string): string {
+    if (!xml.includes("<mxfile")) return xml
+    const doc = new DOMParser().parseFromString(xml, "text/xml")
+    const model = doc.querySelector("diagram")?.querySelector("mxGraphModel")
+    return model ? formatXML(new XMLSerializer().serializeToString(model)) : xml
+}
+
 export default function ChatPanel({
     isVisible,
     onToggleVisibility,
@@ -120,8 +110,9 @@ export default function ChatPanel({
         loadDiagram: onDisplayChart,
         handleExport: onExport,
         handleExportWithoutHistory,
-        resolverRef,
+        exportResolversRef,
         chartXML,
+        chartXMLRef: liveChartXMLRef,
         latestSvg,
         clearDiagram,
         getThumbnailSvg,
@@ -137,21 +128,15 @@ export default function ChatPanel({
     const urlSessionId = searchParams.get("session")
 
     const onFetchChart = (saveToHistory = true) => {
+        // Waits for the reply to its own export, by its tag
+        const tag = saveToHistory ? onExport() : handleExportWithoutHistory()
         return Promise.race([
             new Promise<string>((resolve) => {
-                resolverRef.current = resolve
-                if (saveToHistory) {
-                    onExport()
-                } else {
-                    handleExportWithoutHistory()
-                }
+                if (tag) exportResolversRef.current[tag] = resolve
             }),
             new Promise<string>((_, reject) => {
-                const currentResolver = resolverRef.current
                 setTimeout(() => {
-                    if (resolverRef.current === currentResolver) {
-                        resolverRef.current = null
-                    }
+                    delete exportResolversRef.current[tag]
                     reject(new Error("Chart export timed out after 10 seconds"))
                 }, 10000)
             }),
@@ -178,6 +163,7 @@ export default function ChatPanel({
     const [minimalStyle, setMinimalStyle] = useState(false)
     const [vlmValidationEnabled, setVlmValidationEnabled] = useState(false)
     const [customSystemMessage, setCustomSystemMessage] = useState("")
+    const [maxOutputTokens, setMaxOutputTokens] = useState("")
     const [shouldFocusInput, setShouldFocusInput] = useState(false)
 
     // Restore input from sessionStorage on mount (when ChatPanel remounts due to key change)
@@ -204,6 +190,14 @@ export default function ChatPanel({
         }
     }, [])
 
+    // Load output token budget from localStorage on mount
+    useEffect(() => {
+        const stored = localStorage.getItem(STORAGE_KEYS.maxOutputTokens)
+        if (stored !== null) {
+            setMaxOutputTokens(stored)
+        }
+    }, [])
+
     // Check config on mount
     useEffect(() => {
         fetch(getApiEndpoint("/api/config"))
@@ -281,8 +275,16 @@ export default function ChatPanel({
     // Persist processed tool call IDs so collapsing the chat doesn't replay old tool outputs
     const processedToolCallsRef = useRef<Set<string>>(new Set())
 
-    // Store original XML for edit_diagram streaming - shared between streaming preview and tool handler
-    // Key: toolCallId, Value: original XML before any operations applied
+    // Set by Stop until the user sends the next message
+    const stoppedRef = useRef(false)
+    const preparingSendRef = useRef(false)
+    // Presses of Stop: a check that began before one still knows of it after
+    // the next message clears stoppedRef
+    const stopCountRef = useRef(0)
+
+    // Store original XML for display_diagram and edit_diagram streaming -
+    // shared between streaming preview and tool handler
+    // Key: toolCallId, Value: XML before the call's preview was drawn
     const editDiagramOriginalXmlRef = useRef<Map<string, string>>(new Map())
 
     // Debounce timeout for localStorage writes (prevents blocking during streaming)
@@ -320,28 +322,28 @@ export default function ChatPanel({
         localStorage.setItem(STORAGE_KEYS.customSystemMessage, value)
     }, [])
 
-    // Ref to store the sendMessage function for use in callbacks
-    const sendMessageRef = useRef<typeof sendMessage | null>(null)
-
-    // Callback to improve diagram with validation suggestions
-    const handleImproveWithSuggestions = useCallback((feedback: string) => {
-        if (sendMessageRef.current) {
-            // Send the feedback as a new user message to trigger regeneration
-            sendMessageRef.current({
-                role: "user",
-                parts: [{ type: "text", text: feedback }],
-            })
-        }
+    // Handler for output token budget change (empty string = use server default)
+    const handleMaxOutputTokensChange = useCallback((value: string) => {
+        const digitsOnly = value.replace(/\D/g, "")
+        setMaxOutputTokens(digitsOnly)
+        localStorage.setItem(STORAGE_KEYS.maxOutputTokens, digitsOnly)
     }, [])
 
-    // VLM validation hook using AI SDK's useObject
-    const { validateWithFallback } = useValidateDiagram()
+    // Failed VLM validations in the current user turn (reset on user action)
+    const validationRetryCountRef = useRef(0)
 
-    // Diagram tool handlers (edit_diagram, restructure_diagram, cached replay)
+    // VLM validation hook using AI SDK's useObject
+    const { validateWithFallback, cancel: cancelValidation } =
+        useValidateDiagram()
+
+    // Diagram tool handlers (display_diagram, edit_diagram, append_diagram)
     const { handleToolCall } = useDiagramToolHandlers({
         partialXmlRef,
         editDiagramOriginalXmlRef,
-        chartXMLRef,
+        processedToolCallsRef,
+        validationRetryCountRef,
+        // A preview undone just before the tool call is in this one already
+        chartXMLRef: liveChartXMLRef,
         onDisplayChart,
         onFetchChart,
         onExport,
@@ -349,6 +351,11 @@ export default function ChatPanel({
         validateDiagram: validateWithFallback,
         enableVlmValidation: vlmValidationEnabled,
         sessionId,
+        watchStop: () => {
+            const stopsBefore = stopCountRef.current
+            return () =>
+                stoppedRef.current || stopCountRef.current !== stopsBefore
+        },
         onValidationStateChange: handleValidationStateChange,
     })
 
@@ -367,96 +374,107 @@ export default function ChatPanel({
         onToolCall: async ({ toolCall }) => {
             await handleToolCall({ toolCall }, addToolOutput)
         },
+        onFinish: ({ message, isAbort, isError }) => {
+            // Stopped or failed: tool calls still streaming never reach the
+            // tool handler. Mark them handled so a later render of the
+            // stream does not draw their preview again.
+            if (!isAbort && !isError) return
+            for (const part of message.parts as any[]) {
+                if (part.state === "input-streaming" && part.toolCallId) {
+                    processedToolCallsRef.current.add(part.toolCallId)
+                }
+            }
+        },
         onError: (error) => {
-            // Handle server-side quota limit (429 response)
-            // AI SDK puts the full response body in error.message for non-OK responses
+            // A diagram still streaming when the request failed never
+            // reaches the tool handler: undo its preview. Only previews not
+            // handled yet are stored, and the first one holds the diagram
+            // before any of them.
+            const [originalXml] = editDiagramOriginalXmlRef.current.values()
+            if (originalXml) onDisplayChart(originalXml, true)
+            editDiagramOriginalXmlRef.current.clear()
+
+            // Server errors are JSON: a quota limit ({type: request, token or
+            // tpm}), a provider error ({type: "provider", code, message}) or
+            // {error}. The SDK puts the response body in error.message.
+            let data: any = null
             try {
-                const data = JSON.parse(error.message)
-                if (data.type === "request") {
-                    quotaManager.showQuotaLimitToast(data.used, data.limit)
-                    return
-                }
-                if (data.type === "token") {
-                    quotaManager.showTokenLimitToast(data.used, data.limit)
-                    return
-                }
-                if (data.type === "tpm") {
-                    quotaManager.showTPMLimitToast(data.limit)
-                    return
-                }
+                data = JSON.parse(error.message)
             } catch {
-                // Not JSON, fall through to string matching for backwards compatibility
+                // Plain text, e.g. a network failure in the browser
             }
-
-            // Fallback to string matching
-            if (error.message.includes("Daily request limit")) {
-                quotaManager.showQuotaLimitToast()
+            if (data?.type === "request") {
+                quotaManager.showQuotaLimitToast(data.used, data.limit)
                 return
             }
-            if (error.message.includes("Daily token limit")) {
-                quotaManager.showTokenLimitToast()
+            if (data?.type === "token") {
+                quotaManager.showTokenLimitToast(data.used, data.limit)
                 return
             }
-            if (
-                error.message.includes("Rate limit exceeded") ||
-                error.message.includes("tokens per minute")
-            ) {
-                quotaManager.showTPMLimitToast()
+            if (data?.type === "tpm") {
+                quotaManager.showTPMLimitToast(data.limit)
                 return
             }
 
+            const isAccessCodeError = String(
+                data?.error ?? error.message,
+            ).includes("Invalid or missing access code")
             // Silence access code error in console since it's handled by UI
-            if (!error.message.includes("Invalid or missing access code")) {
-                console.error("Chat error:", error)
-            }
+            if (!isAccessCodeError) console.error("Chat error:", error)
 
-            // Translate technical errors into user-friendly messages
-            // The server now handles detailed error messages, so we can display them directly.
-            // But we still handle connection/network errors that happen before reaching the server.
-            let friendlyMessage = error.message
-
-            // Simple check for network errors if message is generic
-            if (friendlyMessage === "Failed to fetch") {
-                friendlyMessage = "Network error. Please check your connection."
-            }
-
-            // Truncated tool input error (model output limit too low)
-            if (friendlyMessage.includes("toolUse.input is invalid")) {
-                friendlyMessage =
-                    "Output was truncated before the diagram could be generated. Try a simpler request or increase the maxOutputLength."
-            }
-
-            // Translate image not supported error
-            if (
-                friendlyMessage.includes("image content block") ||
-                friendlyMessage.toLowerCase().includes("image_url")
-            ) {
-                friendlyMessage = "This model doesn't support image input."
+            // A hint the user can act on, then the provider's own words
+            let text: string = error.message
+            let openModelConfig = false
+            if (data?.type === "provider") {
+                const hints = dict.errors.llm as Record<string, string>
+                const hint = hints[data.code]
+                text =
+                    hint && data.message
+                        ? `${hint}\n\n${data.message}`
+                        : hint || data.message
+                openModelConfig = [
+                    "invalid_api_key",
+                    "forbidden",
+                    "model_not_found",
+                    "server_key_forbidden",
+                ].includes(data.code)
+            } else if (typeof data?.error === "string") {
+                text = data.error
+            } else if (error.message === "Failed to fetch") {
+                text = dict.errors.networkError
             }
 
             // Add system message for error so it can be cleared
-            setMessages((currentMessages) => {
-                const errorMessage = {
+            setMessages((currentMessages) => [
+                ...currentMessages,
+                {
                     id: `error-${Date.now()}`,
                     role: "system" as const,
-                    content: friendlyMessage,
-                    parts: [{ type: "text" as const, text: friendlyMessage }],
-                }
-                return [...currentMessages, errorMessage]
-            })
+                    content: text,
+                    parts: [{ type: "text" as const, text }],
+                    // The message shows a button that opens model settings
+                    ...(openModelConfig && {
+                        metadata: { openModelConfig: true },
+                    }),
+                },
+            ])
 
-            if (error.message.includes("Invalid or missing access code")) {
+            if (isAccessCodeError) {
                 // Show settings dialog to help user fix it
                 setShowSettingsDialog(true)
             }
         },
-        onFinish: () => {},
+        // Re-render streamed messages at most every 150 ms. The streaming
+        // diagram preview draws on each update, so this also limits redraws
+        experimental_throttle: 150,
         sendAutomaticallyWhen: ({ messages }) => {
+            // The user stopped: a tool result that arrives later (a VLM
+            // check still running) must not start a new request
+            if (stoppedRef.current) return false
+
             const isInContinuationMode = partialXmlRef.current.length > 0
 
-            const shouldRetry = hasToolErrors(
-                messages as unknown as ChatMessage[],
-            )
+            const shouldRetry = hasToolErrors(messages)
 
             if (!shouldRetry) {
                 // No error, reset retry count and clear state
@@ -502,11 +520,6 @@ export default function ChatPanel({
         },
     })
 
-    // Store sendMessage in ref for use in callbacks (like handleImproveWithSuggestions)
-    useEffect(() => {
-        sendMessageRef.current = sendMessage
-    }, [sendMessage])
-
     // Ref to track latest messages for unload persistence
     const messagesRef = useRef(messages)
     useEffect(() => {
@@ -515,6 +528,10 @@ export default function ChatPanel({
 
     // Track last synced session ID to detect external changes (e.g., URL back/forward)
     const lastSyncedSessionIdRef = useRef<string | null>(null)
+    // Message arrays of our own saves. A session holding one of them was
+    // created by our own save, so it must not be treated as an external
+    // switch (with two saves of a new chat at once, the first creates it).
+    const savedMessagesRef = useRef(new WeakSet<object>())
 
     // Helper: Sync UI state with session data (eliminates duplication)
     // Track message IDs that are being loaded from session (to skip animations/scroll)
@@ -581,8 +598,10 @@ export default function ChatPanel({
                     thumbnailDataUrl = latestSvgRef.current
                 }
             }
+            const messages = sanitizeMessages(messagesRef.current)
+            savedMessagesRef.current.add(messages)
             return {
-                messages: sanitizeMessages(messagesRef.current),
+                messages,
                 xmlSnapshots: Array.from(xmlSnapshotsRef.current.entries()),
                 diagramXml: currentDiagramXml,
                 thumbnailDataUrl,
@@ -635,8 +654,13 @@ export default function ChatPanel({
         // Skip if session ID hasn't changed (our own saves don't change the ID)
         if (newSessionId === lastSyncedSessionIdRef.current) return
 
+        // Our own save created this session; the UI already shows its content
+        const isOwnNewSession =
+            !!newSession && savedMessagesRef.current.has(newSession.messages)
+
         // Update last synced ID
         lastSyncedSessionIdRef.current = newSessionId
+        if (isOwnNewSession) return
 
         // Sync UI with new session
         if (newSession) {
@@ -659,6 +683,8 @@ export default function ChatPanel({
         isAvailable: sessionIsAvailable,
         currentSessionId,
         saveCurrentSession,
+        getChatGeneration,
+        getSaveTicket,
     } = sessionManager
 
     // Use ref for saveCurrentSession to avoid infinite loop
@@ -683,26 +709,30 @@ export default function ChatPanel({
             clearTimeout(localStorageDebounceRef.current)
         }
 
-        // Capture current session ID at schedule time to verify at save time
-        const scheduledForSessionId = currentSessionId
+        // Capture the chat on screen at schedule time; the save is dropped
+        // if another chat is on screen by the time it runs
+        const scheduledForChat = getChatGeneration()
         // Capture whether there's a REAL diagram NOW (not just empty template)
         const hasDiagramNow = isRealDiagram(chartXMLRef.current)
         // Check if this session was just loaded without a diagram
         const isNodiagramSession =
-            justLoadedSessionIdRef.current === scheduledForSessionId
+            justLoadedSessionIdRef.current === currentSessionId
 
         // Debounce: save after 1 second of no changes
         localStorageDebounceRef.current = setTimeout(async () => {
             try {
                 if (messages.length > 0 || hasDiagramNow) {
+                    // Taken before the data is read, for the chat it was
+                    // scheduled for
+                    const ticket = {
+                        ...getSaveTicket(),
+                        generation: scheduledForChat,
+                    }
                     const sessionData = await buildSessionData({
                         // Only capture thumbnail if there was a diagram AND this isn't a no-diagram session
                         withThumbnail: hasDiagramNow && !isNodiagramSession,
                     })
-                    await saveCurrentSessionRef.current(
-                        sessionData,
-                        scheduledForSessionId,
-                    )
+                    await saveCurrentSessionRef.current(sessionData, ticket)
                 }
             } catch (error) {
                 console.error("Failed to save session:", error)
@@ -721,6 +751,8 @@ export default function ChatPanel({
         status,
         sessionIsAvailable,
         currentSessionId,
+        getChatGeneration,
+        getSaveTicket,
         buildSessionData,
     ])
 
@@ -753,10 +785,11 @@ export default function ChatPanel({
                 try {
                     // Attempt to save session - browser may not wait for completion
                     // Skip thumbnail capture as it may not complete in time
+                    const ticket = sessionManager.getSaveTicket()
                     const sessionData = await buildSessionData({
                         withThumbnail: false,
                     })
-                    await sessionManager.saveCurrentSession(sessionData)
+                    await sessionManager.saveCurrentSession(sessionData, ticket)
                 } catch (error) {
                     console.error(
                         "Failed to save session on visibility change:",
@@ -774,19 +807,27 @@ export default function ChatPanel({
             )
     }, [sessionManager, buildSessionData])
 
-    const onFormSubmit = async (e: React.FormEvent<HTMLFormElement>) => {
-        e.preventDefault()
+    const submitInput = async () => {
         const isProcessing = status === "streaming" || status === "submitted"
-        if (input.trim() && !isProcessing) {
-            // Check if input matches a cached example (only when no messages yet)
-            if (messages.length === 0) {
+        // Attachments still extracting have no text yet. Template sends call
+        // requestSubmit() and skip the disabled send button, so check here too.
+        const isExtracting =
+            files.some((f) => pdfData.get(f)?.isExtracting) ||
+            Array.from(urlData.values()).some((d) => d.isExtracting)
+        if (input.trim() && !isProcessing && !isExtracting) {
+            // Check if input matches a cached example (only when no messages
+            // yet and the canvas is empty, same rule as the server)
+            if (messages.length === 0 && !hasCells(chartXMLRef.current || "")) {
+                // Pass the file name so a user's own file never matches an example
                 const cached = findCachedResponse(
                     input.trim(),
                     files.length > 0,
+                    files.length === 1 ? files[0].name : undefined,
                 )
                 if (cached) {
-                    // Add user message and fake assistant response to messages
-                    // The chat-message-display useEffect will handle displaying the diagram
+                    // Add the user message and a finished display_diagram
+                    // answer, and load its diagram here: these messages never
+                    // reach the tool handler
                     const toolCallId = `cached-${Date.now()}`
 
                     // Build user message text including any file content
@@ -818,6 +859,16 @@ export default function ChatPanel({
                             ],
                         },
                     ] as any)
+                    // Snapshot the canvas before the example so editing this message works
+                    xmlSnapshotsRef.current.set(
+                        0,
+                        chartXMLRef.current || BLANK_MXFILE,
+                    )
+                    const prepared = prepareNewDiagram(cached.xml, {
+                        pageId: "page-1",
+                        pageName: "Page-1",
+                    })
+                    if (prepared.ok) onDisplayChart(prepared.xml, true)
                     setInput("")
                     sessionStorage.removeItem(SESSION_STORAGE_INPUT_KEY)
                     setFiles([])
@@ -827,9 +878,6 @@ export default function ChatPanel({
             }
 
             try {
-                let chartXml = await onFetchChart()
-                chartXml = formatXML(chartXml)
-
                 // Build user text by concatenating input with pre-extracted text
                 // (Backend only reads first text part, so we must combine them)
                 const parts: any[] = []
@@ -844,48 +892,102 @@ export default function ChatPanel({
                 // Add the combined text as the first part
                 parts.unshift({ type: "text", text: userText })
 
-                // Get previous XML from the last snapshot (before this message)
-                const snapshotKeys = Array.from(
-                    xmlSnapshotsRef.current.keys(),
-                ).sort((a, b) => b - a)
-                const previousXml =
-                    snapshotKeys.length > 0
-                        ? xmlSnapshotsRef.current.get(snapshotKeys[0]) || ""
-                        : ""
-
-                // Save XML snapshot for this message (will be at index = current messages.length)
-                const messageIndex = messages.length
-                xmlSnapshotsRef.current.set(messageIndex, chartXml)
-
-                sendChatMessage(parts, chartXml, previousXml, sessionId)
-
-                // Token count is tracked in onFinish with actual server usage
-                setInput("")
-                sessionStorage.removeItem(SESSION_STORAGE_INPUT_KEY)
-                setFiles([])
-                setUrlData(new Map())
+                await sendWithCurrentDiagram(parts, () => {
+                    setInput("")
+                    sessionStorage.removeItem(SESSION_STORAGE_INPUT_KEY)
+                    setFiles([])
+                    setUrlData(new Map())
+                })
             } catch (error) {
                 console.error("Error fetching chart data:", error)
+                toast.error(dict.errors.failedToExport)
             }
         }
     }
 
+    const onFormSubmit = async (e: React.FormEvent<HTMLFormElement>) => {
+        e.preventDefault()
+        // While a send is prepared (attachments read, diagram exported) the
+        // status is still "ready": a second Enter or click would send the
+        // message again
+        if (preparingSendRef.current) return
+        preparingSendRef.current = true
+        try {
+            await submitInput()
+        } finally {
+            preparingSendRef.current = false
+        }
+    }
+
+    // Export the current diagram, snapshot it for this message, and send.
+    // onSent runs right after sending, so the input empties as the message
+    // shows in the chat
+    const sendWithCurrentDiagram = async (
+        parts: any[],
+        onSent?: () => void,
+    ) => {
+        const chartXml = formatXML(await onFetchChart())
+        const previousXml = getPreviousXml(messages.length)
+
+        // Snapshot the full multi-page document (kept fresh by autosave) so
+        // regenerate/edit can restore every page; the model gets page 1 only
+        xmlSnapshotsRef.current.set(
+            messages.length,
+            chartXMLRef.current || chartXml,
+        )
+
+        sendChatMessage(parts, chartXml, previousXml, sessionId)
+        onSent?.()
+    }
+
+    // Send VLM validation feedback as a new user message through the normal send path
+    const handleImproveWithSuggestions = async (feedback: string) => {
+        if (status === "streaming" || status === "submitted") return
+        try {
+            await sendWithCurrentDiagram([{ type: "text", text: feedback }])
+        } catch (error) {
+            console.error("Error fetching chart data:", error)
+            toast.error(dict.errors.failedToExport)
+        }
+    }
+
+    // The current chat could not be saved (storage full). The list where
+    // old chats can be deleted shows only in an empty chat, so let the user
+    // go on without saving. It replaces the plain message, and has its own
+    // id so a later failed auto-save does not take its button away.
+    const offerToContinueUnsaved = useCallback(
+        (proceed: () => void) => {
+            toast.dismiss("session-save-failed")
+            toast.error(dict.errors.sessionSaveFailedLeave, {
+                id: "session-save-leave",
+                duration: 15000,
+                action: {
+                    label: dict.errors.continueWithoutSaving,
+                    onClick: proceed,
+                },
+            })
+        },
+        [dict],
+    )
+
+    // A new turn makes the offer stale: going on would clear the chat while
+    // the answer streams in
+    useEffect(() => {
+        if (status === "submitted" || status === "streaming") {
+            toast.dismiss("session-save-leave")
+        }
+    }, [status])
+
     // Handle session switching from history dropdown
     const handleSelectSession = useCallback(
         async (sessionId: string) => {
             if (!sessionManager.isAvailable) return
 
-            // Save current session before switching
-            if (messages.length > 0) {
-                const sessionData = await buildSessionData({
-                    withThumbnail: true,
-                })
-                await sessionManager.saveCurrentSession(sessionData)
-            }
-
             // Switch to selected session
-            const sessionData = await sessionManager.switchSession(sessionId)
-            if (sessionData) {
+            const open = async () => {
+                const sessionData =
+                    await sessionManager.switchSession(sessionId)
+                if (!sessionData) return
                 const hasRealDiagram = isRealDiagram(sessionData.diagramXml)
                 justLoadedSessionRef.current = true
 
@@ -903,8 +1005,37 @@ export default function ChatPanel({
                 syncUIWithSession(sessionData)
                 router.replace(`?session=${sessionId}`, { scroll: false })
             }
+
+            // Save current session before switching (also a diagram drawn
+            // without messages); if that failed (storage full), stay on it
+            // unless the user goes on without saving it
+            if (messages.length > 0 || isRealDiagram(chartXMLRef.current)) {
+                // Of the chat on screen now, also if another one comes on
+                // screen while the thumbnail is taken
+                const ticket = sessionManager.getSaveTicket()
+                const sessionData = await buildSessionData({
+                    withThumbnail: true,
+                })
+                if (
+                    !(await sessionManager.saveCurrentSession(
+                        sessionData,
+                        ticket,
+                    ))
+                ) {
+                    offerToContinueUnsaved(open)
+                    return
+                }
+            }
+            await open()
         },
-        [sessionManager, messages, buildSessionData, syncUIWithSession, router],
+        [
+            sessionManager,
+            messages,
+            buildSessionData,
+            syncUIWithSession,
+            router,
+            offerToContinueUnsaved,
+        ],
     )
 
     // Handle session deletion from history dropdown
@@ -922,15 +1053,7 @@ export default function ChatPanel({
         [sessionManager, syncUIWithSession, router, pathname],
     )
 
-    const handleNewChat = useCallback(async () => {
-        // Save current session before creating new one
-        if (sessionManager.isAvailable && messages.length > 0) {
-            const sessionData = await buildSessionData({ withThumbnail: true })
-            await sessionManager.saveCurrentSession(sessionData)
-            // Refresh sessions list to ensure dropdown shows the saved session
-            await sessionManager.refreshSessions()
-        }
-
+    const startNewChat = useCallback(() => {
         // Clear session manager state BEFORE clearing URL to prevent race condition
         // (otherwise the URL update effect would restore the old session URL)
         sessionManager.clearCurrentSession()
@@ -962,21 +1085,47 @@ export default function ChatPanel({
         setMessages,
         setSessionId,
         sessionManager,
-        messages,
         router,
         dict.dialogs.clearSuccess,
-        buildSessionData,
         setDiagramHistory,
         pathname,
     ])
 
+    const handleNewChat = useCallback(async () => {
+        // Save current session before creating new one (also a diagram
+        // drawn without messages)
+        if (
+            sessionManager.isAvailable &&
+            (messages.length > 0 || isRealDiagram(chartXMLRef.current))
+        ) {
+            const ticket = sessionManager.getSaveTicket()
+            const sessionData = await buildSessionData({ withThumbnail: true })
+            // Not saved (storage full): keep the chat on screen, unless the
+            // user goes on without saving it
+            if (
+                !(await sessionManager.saveCurrentSession(sessionData, ticket))
+            ) {
+                offerToContinueUnsaved(startNewChat)
+                return
+            }
+            // Refresh sessions list to ensure dropdown shows the saved session
+            await sessionManager.refreshSessions()
+        }
+        startNewChat()
+    }, [
+        sessionManager,
+        messages,
+        buildSessionData,
+        offerToContinueUnsaved,
+        startNewChat,
+    ])
+
     // Handle sending a template directly (called from TemplatePanel)
     const handleSendTemplate = useCallback(
         async (template: { prompt: string }) => {
+            // Keep attachments: they are sent along with the template prompt
             flushSync(() => {
                 setInput(template.prompt)
-                setFiles([])
-                setUrlData(new Map())
             })
 
             const formElement = document.getElementById(
@@ -986,7 +1135,7 @@ export default function ChatPanel({
                 formElement.requestSubmit()
             }
         },
-        [setInput, setFiles, setUrlData],
+        [setInput],
     )
 
     const handleInputChange = (
@@ -1001,13 +1150,15 @@ export default function ChatPanel({
     }
 
     // Helper functions for message actions (regenerate/edit)
-    // Extract previous XML snapshot before a given message index
+    // Extract previous XML snapshot (first page, as sent to the model) before a given message index
     const getPreviousXml = (beforeIndex: number): string => {
         const snapshotKeys = Array.from(xmlSnapshotsRef.current.keys())
             .filter((k) => k < beforeIndex)
             .sort((a, b) => b - a)
         return snapshotKeys.length > 0
-            ? xmlSnapshotsRef.current.get(snapshotKeys[0]) || ""
+            ? getFirstPageXml(
+                  xmlSnapshotsRef.current.get(snapshotKeys[0]) || "",
+              )
             : ""
     }
 
@@ -1028,11 +1179,19 @@ export default function ChatPanel({
 
     // Handle stop button click
     const handleStop = useCallback(() => {
+        stoppedRef.current = true
+        stopCountRef.current++
+        // A running screenshot check holds up the chat (the SDK waits for
+        // the tool handler): end it, so the call gets its result now
+        cancelValidation()
         const lastMessage = messages[messages.length - 1]
+        // Calls the tool handler already took can still show as streaming:
+        // the messages update at most every 150 ms (useChat throttle)
         const toolParts = lastMessage?.parts?.filter(
             (part: any) =>
                 part.type?.startsWith("tool-") &&
-                part.state === "input-streaming",
+                part.state === "input-streaming" &&
+                !processedToolCallsRef.current.has(part.toolCallId),
         )
 
         toolParts?.forEach((part: any) => {
@@ -1047,7 +1206,7 @@ export default function ChatPanel({
         })
 
         stop()
-    }, [messages, addToolOutput, stop])
+    }, [messages, addToolOutput, stop, cancelValidation])
 
     // Send chat message with headers
     const sendChatMessage = (
@@ -1059,7 +1218,9 @@ export default function ChatPanel({
         // Reset all retry/continuation state on user-initiated message
         autoRetryCountRef.current = 0
         continuationRetryCountRef.current = 0
+        validationRetryCountRef.current = 0
         partialXmlRef.current = ""
+        stoppedRef.current = false
 
         const config = getSelectedAIConfig()
 
@@ -1104,6 +1265,9 @@ export default function ChatPanel({
                     ...(minimalStyle && {
                         "x-minimal-style": "true",
                     }),
+                    ...(maxOutputTokens && {
+                        "x-max-output-tokens": maxOutputTokens,
+                    }),
                 },
             },
         )
@@ -1204,7 +1368,12 @@ export default function ChatPanel({
         })
 
         // Now send the message after state is guaranteed to be updated
-        sendChatMessage(userParts, savedXml, previousXml, sessionId)
+        sendChatMessage(
+            userParts,
+            getFirstPageXml(savedXml),
+            previousXml,
+            sessionId,
+        )
     }
 
     const handleEditMessage = async (messageIndex: number, newText: string) => {
@@ -1231,10 +1400,13 @@ export default function ChatPanel({
         // Clean up snapshots for messages after the user message (they will be removed)
         cleanupSnapshotsAfter(messageIndex)
 
-        // Create new parts with updated text
+        // Create new parts with updated text. The edit box only shows the typed
+        // text, so keep the appended PDF/file/URL content
         const newParts = message.parts?.map((part: any) => {
             if (part.type === "text") {
-                return { ...part, text: newText }
+                const appended =
+                    part.text.match(APPENDED_FILE_SECTIONS_PATTERN)?.[0] ?? ""
+                return { ...part, text: newText + appended }
             }
             return part
         }) || [{ type: "text", text: newText }]
@@ -1247,7 +1419,12 @@ export default function ChatPanel({
         })
 
         // Now send the edited message after state is guaranteed to be updated
-        sendChatMessage(newParts, savedXml, previousXml, sessionId)
+        sendChatMessage(
+            newParts,
+            getFirstPageXml(savedXml),
+            previousXml,
+            sessionId,
+        )
     }
 
     // Collapsed view (desktop only)
@@ -1376,6 +1553,7 @@ export default function ChatPanel({
             {/* Messages */}
             <main className="flex-1 w-full overflow-hidden">
                 <ChatMessageDisplay
+                    onOpenModelConfig={() => setShowModelConfigDialog(true)}
                     messages={messages}
                     setInput={setInput}
                     setFiles={handleFileChange}
@@ -1448,6 +1626,8 @@ export default function ChatPanel({
                 onVlmValidationChange={handleVlmValidationChange}
                 customSystemMessage={customSystemMessage}
                 onCustomSystemMessageChange={handleCustomSystemMessageChange}
+                maxOutputTokens={maxOutputTokens}
+                onMaxOutputTokensChange={handleMaxOutputTokensChange}
                 onOpenModelConfig={() => setShowModelConfigDialog(true)}
             />
 

components/chat/ChatLobby.tsx

@@ -194,6 +194,8 @@ export function ChatLobby({
                                     className="group w-full flex items-center gap-3 p-3 rounded-xl border border-border/60 bg-card hover:bg-accent/50 hover:border-primary/30 transition-all duration-200 cursor-pointer text-left"
                                     onClick={() => onSelectSession(session.id)}
                                     onKeyDown={(e) => {
+                                        // Ignore keys bubbling up from the delete button
+                                        if (e.target !== e.currentTarget) return
                                         if (
                                             e.key === "Enter" ||
                                             e.key === " "

components/chat/TemplateCreateDialog.tsx

@@ -55,6 +55,9 @@ export function TemplateCreateDialog({
 
     const handleSubmit = async (e: React.FormEvent) => {
         e.preventDefault()
+        // React submit events bubble through the portal; keep them away from
+        // the chat form this dialog may be rendered in
+        e.stopPropagation()
 
         const trimmedPrompt = prompt.trim()
         if (!trimmedPrompt) {

components/chat/TemplateEditDialog.tsx

@@ -39,16 +39,16 @@ export function TemplateEditDialog({
     const [isSubmitting, setIsSubmitting] = useState(false)
     const [error, setError] = useState<string | null>(null)
 
-    // Populate form when template changes
+    // Populate form each time the dialog opens, dropping any cancelled edits
     useEffect(() => {
-        if (template) {
+        if (open && template) {
             setTitle(template.title || "")
             setDescription(template.description || "")
             setPrompt(template.prompt || "")
             setPinned(template.pinned || false)
             setError(null)
         }
-    }, [template])
+    }, [open, template])
 
     const handleOpenChange = (newOpen: boolean) => {
         if (!newOpen) {
@@ -59,6 +59,9 @@ export function TemplateEditDialog({
 
     const handleSubmit = async (e: React.FormEvent) => {
         e.preventDefault()
+        // React submit events bubble through the portal; keep them away from
+        // any form this dialog may be rendered in
+        e.stopPropagation()
 
         if (!template) return
 

components/chat/TemplatePanel.tsx

@@ -110,6 +110,10 @@ export function TemplatePanel({
 
     useEffect(() => {
         loadTemplates()
+        // Reload when a template is saved elsewhere, e.g. from the chat input
+        window.addEventListener("templatesChanged", loadTemplates)
+        return () =>
+            window.removeEventListener("templatesChanged", loadTemplates)
     }, [loadTemplates])
 
     const handleCreateSuccess = () => {
@@ -220,7 +224,8 @@ export function TemplatePanel({
             const url = URL.createObjectURL(blob)
             const a = document.createElement("a")
             a.href = url
-            a.download = `templates-${new Date().toISOString().split("T")[0]}.json`
+            // Local date as YYYY-MM-DD (toISOString would give UTC)
+            a.download = `templates-${new Date().toLocaleDateString("sv-SE")}.json`
             document.body.appendChild(a)
             a.click()
             document.body.removeChild(a)
@@ -302,6 +307,28 @@ export function TemplatePanel({
         }
     }
 
+    // Shared by the empty state and the list, so import works in both
+    const importInput = (
+        <input
+            ref={fileInputRef}
+            type="file"
+            accept="application/json,.json"
+            onChange={handleImport}
+            className="hidden"
+        />
+    )
+    const importMessageBox = importMessage && (
+        <div
+            className={`text-xs px-3 py-2 rounded-lg ${
+                importMessage.type === "success"
+                    ? "bg-green-100 text-green-800 dark:bg-green-900/30 dark:text-green-400"
+                    : "bg-red-100 text-red-800 dark:bg-red-900/30 dark:text-red-400"
+            }`}
+        >
+            {importMessage.text}
+        </div>
+    )
+
     // Empty state: no templates at all
     if (!loading && templates.length === 0) {
         return (
@@ -332,6 +359,18 @@ export function TemplatePanel({
                         <Plus className="w-4 h-4" />
                         {dict.templates.createFirst}
                     </button>
+                    <button
+                        type="button"
+                        onClick={() => fileInputRef.current?.click()}
+                        className="mt-2 inline-flex items-center gap-1.5 px-3 py-1.5 rounded-md text-xs font-medium text-muted-foreground hover:text-foreground hover:bg-muted transition-colors"
+                    >
+                        <Upload className="w-3.5 h-3.5" />
+                        {dict.templates.importTemplates}
+                    </button>
+                    {importInput}
+                    {importMessageBox && (
+                        <div className="mt-3">{importMessageBox}</div>
+                    )}
 
                     <TemplateCreateDialog
                         open={createDialogOpen}
@@ -389,27 +428,11 @@ export function TemplatePanel({
                         <Upload className="w-3.5 h-3.5" />
                         {dict.templates.importTemplates}
                     </button>
-                    <input
-                        ref={fileInputRef}
-                        type="file"
-                        accept="application/json,.json"
-                        onChange={handleImport}
-                        className="hidden"
-                    />
+                    {importInput}
                 </div>
 
                 {/* Import message */}
-                {importMessage && (
-                    <div
-                        className={`text-xs px-3 py-2 rounded-lg ${
-                            importMessage.type === "success"
-                                ? "bg-green-100 text-green-800 dark:bg-green-900/30 dark:text-green-400"
-                                : "bg-red-100 text-red-800 dark:bg-red-900/30 dark:text-red-400"
-                        }`}
-                    >
-                        {importMessage.text}
-                    </div>
-                )}
+                {importMessageBox}
 
                 <div className="space-y-2">
                     {loading
@@ -447,6 +470,8 @@ export function TemplatePanel({
                                         handleTemplateClick(template)
                                     }
                                     onKeyDown={(e) => {
+                                        // Ignore keys bubbling up from the action buttons
+                                        if (e.target !== e.currentTarget) return
                                         if (
                                             e.key === "Enter" ||
                                             e.key === " "

components/chat/ToolCallCard.tsx

@@ -4,11 +4,7 @@ import { Check, ChevronDown, ChevronUp, Copy, Cpu } from "lucide-react"
 import type { Dispatch, SetStateAction } from "react"
 import { CodeBlock } from "@/components/code-block"
 import { isMxCellXmlComplete } from "@/lib/utils"
-import type {
-    DiagramOperation,
-    StructureOperation,
-    ToolPartLike,
-} from "./types"
+import type { DiagramOperation, ToolPartLike } from "./types"
 
 interface ToolCallCardProps {
     part: ToolPartLike
@@ -23,140 +19,40 @@ interface ToolCallCardProps {
     }
 }
 
-/**
- * Colour an operation by what it does to the diagram: removes, adds, or changes.
- *
- * Takes an unknown rather than a string because this renders DURING streaming: the tool
- * input arrives character by character, so an operation is briefly `{}` or `{"op": "add_c`
- * before it is whole. A missing name is the normal mid-stream state, not an error.
- */
-function opColour(op: unknown): string {
-    if (typeof op !== "string") return "text-muted-foreground"
-    if (op === "delete" || op === "remove" || op === "unlink" || op === "clear")
-        return "text-red-600"
-    if (op.startsWith("add") || op === "link") return "text-green-600"
-    return "text-blue-600"
-}
-
-/**
- * The arguments worth showing beside an operation's name.
- *
- * A whitelist rather than "everything except op and id", because some operations carry a
- * whole nested graph (add_graph's nodes and edges) and dumping that turns one line into a
- * screenful. The excluded keys are summarised instead.
- */
-const SHOWN_KEYS = [
-    "label",
-    "name",
-    "parent",
-    "dir",
-    "class",
-    "role",
-    "group",
-    "shape",
-    "cols",
-    "lanes",
-    "aspect",
-    "source",
-    "target",
-    "title",
-] as const
-
-function summarise(op: StructureOperation | undefined | null): string {
-    if (!op || typeof op !== "object") return ""
-    const parts: string[] = []
-    for (const key of SHOWN_KEYS) {
-        const v = op[key]
-        if (v === undefined || v === null || v === "") continue
-        parts.push(
-            `${key}=${Array.isArray(v) ? v.join("/") : String(v).slice(0, 60)}`,
-        )
-    }
-    // A graph carries its own nodes and edges; report the size, not the contents.
-    const nodes = op.nodes
-    const edges = op.edges
-    if (Array.isArray(nodes))
-        parts.push(
-            `${nodes.length} node${nodes.length === 1 ? "" : "s"}${
-                Array.isArray(edges)
-                    ? `, ${edges.length} edge${edges.length === 1 ? "" : "s"}`
-                    : ""
-            }`,
-        )
-    return parts.join("  ")
-}
-
-/**
- * `restructure_diagram`'s operations: structural steps, not XML patches.
- *
- * Written to survive PARTIAL data. This renders while the tool input is still streaming, so
- * an entry may be `{}`, or `{op: "add_contai"}`, or — because a JSON array is repaired as it
- * arrives — `undefined`. Every field is therefore treated as possibly absent rather than
- * validated up front: dropping incomplete entries would make rows appear and disappear as
- * the text arrives, and asserting on them crashes the whole message.
- */
-function StructureOperationsDisplay({
-    operations,
-}: {
-    operations: StructureOperation[]
-}) {
-    return (
-        <div className="space-y-1">
-            {operations.map((op, index) => (
-                <div
-                    key={`${op?.op ?? "pending"}-${op?.id ?? index}-${index}`}
-                    className="flex items-baseline gap-2 px-2 py-1 rounded bg-background/50 border border-border/40"
-                >
-                    <span
-                        className={`text-[10px] font-medium uppercase tracking-wide shrink-0 ${opColour(op?.op)}`}
-                    >
-                        {op?.op ?? "…"}
-                    </span>
-                    {op?.id && (
-                        <span className="text-xs font-mono text-foreground/80 shrink-0">
-                            {op.id}
-                        </span>
-                    )}
-                    <span className="text-[11px] text-muted-foreground font-mono break-all">
-                        {summarise(op)}
-                    </span>
-                </div>
-            ))}
-        </div>
-    )
-}
-
-/** `edit_diagram`'s operations. Also streamed, so also written for partial entries. */
 function OperationsDisplay({ operations }: { operations: DiagramOperation[] }) {
+    // Streamed or invalid input can hold anything: show only what React can
+    // render (an object in place of a string would crash the whole chat)
+    const shown = operations.filter(
+        (op) => typeof (op as { operation?: unknown })?.operation === "string",
+    )
+    const text = (value: unknown) => (typeof value === "string" ? value : "")
     return (
         <div className="space-y-3">
-            {operations.map((op, index) => (
+            {shown.map((op, index) => (
                 <div
-                    key={`${op?.operation ?? "pending"}-${op?.cell_id ?? index}-${index}`}
+                    key={`${op.operation}-${text(op.cell_id)}-${index}`}
                     className="rounded-lg border border-border/50 overflow-hidden bg-background/50"
                 >
                     <div className="px-3 py-1.5 bg-muted/40 border-b border-border/30 flex items-center gap-2">
                         <span
                             className={`text-[10px] font-medium uppercase tracking-wide ${
-                                op?.operation === "delete"
+                                op.operation === "delete"
                                     ? "text-red-600"
-                                    : op?.operation === "add"
+                                    : op.operation === "add"
                                       ? "text-green-600"
                                       : "text-blue-600"
                             }`}
                         >
-                            {op?.operation ?? "…"}
+                            {op.operation}
+                        </span>
+                        <span className="text-xs text-muted-foreground">
+                            cell_id: {text(op.cell_id)}
                         </span>
-                        {op?.cell_id && (
-                            <span className="text-xs text-muted-foreground">
-                                cell_id: {op.cell_id}
-                            </span>
-                        )}
                     </div>
-                    {op?.new_xml && (
+                    {text(op.new_xml) && (
                         <div className="px-3 py-2">
                             <pre className="text-[11px] font-mono text-foreground/80 bg-muted/30 rounded px-2 py-1.5 overflow-x-auto whitespace-pre-wrap break-all">
-                                {op.new_xml}
+                                {text(op.new_xml)}
                             </pre>
                         </div>
                     )}
@@ -176,7 +72,7 @@ export function ToolCallCard({
     dict,
 }: ToolCallCardProps) {
     const callId = part.toolCallId
-    const { state, input, output } = part
+    const { state, input, output, errorText } = part
     // Default to expanded for all states (user can manually collapse if needed)
     const isExpanded = expandedTools[callId] ?? true
     const toolName = part.type?.replace("tool-", "")
@@ -191,21 +87,27 @@ export function ToolCallCard({
 
     const getToolDisplayName = (name: string) => {
         switch (name) {
-            case "restructure_diagram":
-                return "Build Diagram"
-            case "edit_diagram":
-                return "Edit Diagram"
-            case "search_stencils":
-                return "Find Icons"
-            // Only ever arrives from the server's cache-hit path now; the model cannot
-            // call it. See createCachedStreamResponse in app/api/chat/route.ts.
             case "display_diagram":
                 return "Generate Diagram"
+            case "edit_diagram":
+                return "Edit Diagram"
+            case "get_shape_library":
+                return "Get Shape Library"
             default:
                 return name
         }
     }
 
+    // Incomplete XML means the output hit the length limit, unless the user
+    // stopped the generation themselves. Without an input the JSON was
+    // broken (the server repairs JSON cut short by the limit).
+    const isTruncated =
+        state === "output-error" &&
+        errorText !== "Stopped by user" &&
+        (toolName === "display_diagram" || toolName === "append_diagram") &&
+        typeof input?.xml === "string" &&
+        !isMxCellXmlComplete(input.xml)
+
     const handleCopy = () => {
         let textToCopy = ""
 
@@ -219,6 +121,14 @@ export function ToolCallCard({
             }
         }
 
+        if (
+            output &&
+            toolName === "get_shape_library" &&
+            typeof output === "string"
+        ) {
+            textToCopy = output
+        }
+
         if (textToCopy) {
             onCopy(callId, textToCopy, true)
         }
@@ -267,23 +177,15 @@ export function ToolCallCard({
                         </>
                     )}
                     {state === "output-error" &&
-                        (() => {
-                            // Truncation only applies to a tool that streams raw XML, which
-                            // is now just the cached-answer replay. The engine tools send
-                            // structured operations, so a failure there is a real error.
-                            const isTruncated =
-                                toolName === "display_diagram" &&
-                                !isMxCellXmlComplete(input?.xml)
-                            return isTruncated ? (
-                                <span className="text-xs font-medium text-yellow-600 bg-yellow-50 px-2 py-0.5 rounded-full">
-                                    Truncated
-                                </span>
-                            ) : (
-                                <span className="text-xs font-medium text-red-600 bg-red-50 px-2 py-0.5 rounded-full">
-                                    Error
-                                </span>
-                            )
-                        })()}
+                        (isTruncated ? (
+                            <span className="text-xs font-medium text-yellow-600 bg-yellow-50 px-2 py-0.5 rounded-full">
+                                Truncated
+                            </span>
+                        ) : (
+                            <span className="text-xs font-medium text-red-600 bg-red-50 px-2 py-0.5 rounded-full">
+                                Error
+                            </span>
+                        ))}
                     {input && Object.keys(input).length > 0 && (
                         <button
                             type="button"
@@ -321,23 +223,7 @@ export function ToolCallCard({
                     ) : typeof input === "object" &&
                       input.operations &&
                       Array.isArray(input.operations) ? (
-                        // Dispatch by TOOL, not by whether an `operations` key exists: both
-                        // tools call their argument that, but the items have different shapes
-                        // (op/id versus operation/cell_id), and reading one as the other
-                        // printed a row of blank `cell_id:` labels.
-                        toolName === "restructure_diagram" ? (
-                            <StructureOperationsDisplay
-                                operations={
-                                    input.operations as StructureOperation[]
-                                }
-                            />
-                        ) : (
-                            <OperationsDisplay
-                                operations={
-                                    input.operations as DiagramOperation[]
-                                }
-                            />
-                        )
+                        <OperationsDisplay operations={input.operations} />
                     ) : typeof input === "object" &&
                       Object.keys(input).length > 0 ? (
                         <CodeBlock
@@ -347,37 +233,35 @@ export function ToolCallCard({
                     ) : null}
                 </div>
             )}
-            {output &&
-                state === "output-error" &&
-                (() => {
-                    const isTruncated =
-                        toolName === "display_diagram" &&
-                        !isMxCellXmlComplete(input?.xml)
-                    return (
-                        <div
-                            className={`px-4 py-3 border-t border-border/40 text-sm ${isTruncated ? "text-yellow-600" : "text-red-600"}`}
-                        >
-                            {isTruncated
-                                ? "Output truncated due to length limits. Try a simpler request or increase the maxOutputLength."
-                                : output}
-                        </div>
-                    )
-                })()}
-            {/* What the tool actually returned. Worth showing on success, not only on
-                error: restructure_diagram answers with an outline of the structure it
-                built plus any notes about classes it could not honour, and that is the
-                same text the model reads to name ids in its next call. */}
-            {output && state === "output-available" && isExpanded && (
-                <div className="px-4 py-3 border-t border-border/40">
-                    <pre className="text-[11px] font-mono text-muted-foreground bg-muted/40 rounded-md p-2 overflow-auto max-h-64 whitespace-pre-wrap break-all">
-                        {typeof output === "string"
-                            ? output.length > 4000
-                                ? `${output.slice(0, 4000)}\n…`
-                                : output
-                            : String(output)}
-                    </pre>
+            {/* AI SDK stores tool errors in errorText */}
+            {state === "output-error" && (errorText || output) && (
+                <div
+                    className={`px-4 py-3 border-t border-border/40 text-sm whitespace-pre-wrap break-words ${isTruncated ? "text-yellow-600" : "text-red-600"}`}
+                >
+                    {isTruncated
+                        ? "Output truncated due to length limits. Try a simpler request or increase Max Output Tokens in Settings."
+                        : (errorText ?? output)}
                 </div>
             )}
+            {/* Show get_shape_library output on success */}
+            {output &&
+                toolName === "get_shape_library" &&
+                state === "output-available" &&
+                isExpanded && (
+                    <div className="px-4 py-3 border-t border-border/40">
+                        <div className="text-xs text-muted-foreground mb-2">
+                            Library loaded (
+                            {typeof output === "string" ? output.length : 0}{" "}
+                            chars)
+                        </div>
+                        <pre className="text-xs bg-muted/50 p-2 rounded-md overflow-auto max-h-32 whitespace-pre-wrap">
+                            {typeof output === "string"
+                                ? output.substring(0, 800) +
+                                  (output.length > 800 ? "\n..." : "")
+                                : String(output)}
+                        </pre>
+                    </div>
+                )}
         </div>
     )
 }

components/chat/types.ts

@@ -1,30 +1,6 @@
-/** An `edit_diagram` operation: a patch against one cell, addressed by its id. */
-export interface DiagramOperation {
-    operation: "update" | "add" | "delete"
-    cell_id: string
-    new_xml?: string
-}
+import type { DiagramOperation } from "@/packages/mcp-server/src/diagram-operations.ts"
 
-/**
- * A `restructure_diagram` operation.
- *
- * Deliberately loose. The engine owns the real schema (lib/diagram-engine/operations.ts)
- * and it has two dozen variants; the card only needs to say WHAT each step did, so it
- * reads the two fields every variant shares and picks a few recognisable extras out of
- * the rest. Mirroring the full union here would mean editing this file every time the
- * engine gains an operation.
- *
- * The field names matter: `op`/`id`, where edit_diagram has `operation`/`cell_id`. Both
- * tools happen to call their argument `operations`, which is what let the card render one
- * as the other and print six blank `cell_id:` lines.
- */
-export interface StructureOperation {
-    op: string
-    id?: string
-    label?: string
-    parent?: string
-    [key: string]: unknown
-}
+export type { DiagramOperation }
 
 export interface ToolPartLike {
     type: string
@@ -32,7 +8,8 @@ export interface ToolPartLike {
     state?: string
     input?: {
         xml?: string
-        operations?: DiagramOperation[] | StructureOperation[]
+        operations?: DiagramOperation[]
     } & Record<string, unknown>
     output?: string
+    errorText?: string
 }

components/dev-xml-simulator.tsx

@@ -2,7 +2,7 @@
 
 import { useEffect, useRef, useState } from "react"
 import { useDictionary } from "@/hooks/use-dictionary"
-import { wrapWithMxFile } from "@/lib/utils"
+import { prepareNewDiagram } from "@/packages/mcp-server/src/new-diagram.ts"
 
 // Dev XML presets for streaming simulator
 const DEV_XML_PRESETS: Record<string, string> = {
@@ -237,8 +237,8 @@ export function DevXmlSimulator({
         })
 
         // Display the final diagram
-        const fullXml = wrapWithMxFile(xml)
-        onDisplayChart(fullXml)
+        const prepared = prepareNewDiagram(xml)
+        if (prepared.ok) onDisplayChart(prepared.xml)
 
         setIsSimulating(false)
     }

components/model-config-dialog.tsx

@@ -4,12 +4,12 @@ import {
     AlertCircle,
     Check,
     ChevronRight,
-    Clock,
     Eye,
     EyeOff,
     Key,
     Loader2,
     Plus,
+    RefreshCw,
     Server,
     Settings2,
     Sparkles,
@@ -34,6 +34,13 @@ import {
     AlertDialogTitle,
 } from "@/components/ui/alert-dialog"
 import { Button } from "@/components/ui/button"
+import {
+    Command,
+    CommandEmpty,
+    CommandInput,
+    CommandItem,
+    CommandList,
+} from "@/components/ui/command"
 import {
     Dialog,
     DialogContent,
@@ -43,6 +50,11 @@ import {
 } from "@/components/ui/dialog"
 import { Input } from "@/components/ui/input"
 import { Label } from "@/components/ui/label"
+import {
+    Popover,
+    PopoverContent,
+    PopoverTrigger,
+} from "@/components/ui/popover"
 import { ScrollArea } from "@/components/ui/scroll-area"
 import {
     Select,
@@ -56,7 +68,13 @@ import { useDictionary } from "@/hooks/use-dictionary"
 import type { UseModelConfigReturn } from "@/hooks/use-model-config"
 import { getApiEndpoint } from "@/lib/base-path"
 import { formatMessage } from "@/lib/i18n/utils"
-import type { ProviderConfig, ProviderName } from "@/lib/types/model-config"
+import type { ListedModel } from "@/lib/provider-models"
+import { STORAGE_KEYS } from "@/lib/storage"
+import type {
+    ModelConfig,
+    ProviderConfig,
+    ProviderName,
+} from "@/lib/types/model-config"
 import { PROVIDER_INFO, SUGGESTED_MODELS } from "@/lib/types/model-config"
 import { cn } from "@/lib/utils"
 
@@ -125,22 +143,31 @@ export function ModelConfigDialog({
     > | null>(null)
     const [deleteConfirmOpen, setDeleteConfirmOpen] = useState(false)
     const [deleteConfirmText, setDeleteConfirmText] = useState("")
-    const [validatingModelIndex, setValidatingModelIndex] = useState<
-        number | null
-    >(null)
+    // Models whose test is running (they are all tested at once)
+    const [validatingModelIds, setValidatingModelIds] = useState<Set<string>>(
+        () => new Set(),
+    )
     const [duplicateError, setDuplicateError] = useState<string>("")
     const [editError, setEditError] = useState<{
         modelId: string
         message: string
     } | null>(null)
-    const [dynamicSuggestedModels, setDynamicSuggestedModels] = useState<
-        Partial<Record<ProviderName, string[]>>
+    // Model ID being typed; written to the config only when valid on blur
+    const [modelIdDraft, setModelIdDraft] = useState<{
+        id: string
+        value: string
+    } | null>(null)
+    // Models fetched from the provider, per provider config
+    const [fetchedModels, setFetchedModels] = useState<
+        Record<string, ListedModel[]>
     >({})
-    const [loadedSuggestedProviders, setLoadedSuggestedProviders] = useState<
-        Partial<Record<ProviderName, boolean>>
-    >({})
-    const [loadingSuggestedProvider, setLoadingSuggestedProvider] =
-        useState<ProviderName | null>(null)
+    const [fetchingModels, setFetchingModels] = useState(false)
+    const [fetchModelsError, setFetchModelsError] = useState("")
+    const [modelPickerOpen, setModelPickerOpen] = useState(false)
+    // models.dev data for hints, loaded with the dialog (it is ~180 KB)
+    const [getModelInfo, setGetModelInfo] = useState<
+        typeof import("@/lib/model-catalog").getModelInfo | null
+    >(null)
 
     const {
         config,
@@ -156,6 +183,35 @@ export function ModelConfigDialog({
     const selectedProvider = config.providers.find(
         (p) => p.id === selectedProviderId,
     )
+    // For requests that finish after the user switched provider or edited
+    // a model id
+    const selectedProviderIdRef = useRef(selectedProviderId)
+    selectedProviderIdRef.current = selectedProviderId
+    const configRef = useRef(config)
+    configRef.current = config
+    // Number of the latest Test click: only that test may reset the busy
+    // state when its credentials changed meanwhile
+    const validationRunRef = useRef(0)
+    // A model list or test result belongs to the credentials it was asked
+    // with; they can change meanwhile, here or in another tab
+    const credentialsOf = (providerId: string) => {
+        const p = configRef.current.providers.find((x) => x.id === providerId)
+        return JSON.stringify([
+            p?.provider,
+            p?.apiKey,
+            p?.baseUrl,
+            p?.awsAccessKeyId,
+            p?.awsSecretAccessKey,
+            p?.awsRegion,
+            p?.awsSessionToken,
+            p?.vertexApiKey,
+        ])
+    }
+
+    // Discard an unfinished model ID edit when the dialog closes
+    useEffect(() => {
+        if (!open) setModelIdDraft(null)
+    }, [open])
 
     // Cleanup validation reset timeout on unmount
     useEffect(() => {
@@ -167,73 +223,83 @@ export function ModelConfigDialog({
     }, [])
 
     useEffect(() => {
-        if (
-            !open ||
-            selectedProvider?.provider !== "aihubmix" ||
-            loadedSuggestedProviders.aihubmix
-        ) {
-            return
-        }
+        if (!open || getModelInfo) return
+        import("@/lib/model-catalog").then((catalog) =>
+            setGetModelInfo(() => catalog.getModelInfo),
+        )
+    }, [open, getModelInfo])
 
-        let cancelled = false
-        setLoadingSuggestedProvider("aihubmix")
-
-        fetch(getApiEndpoint("/api/aihubmix-models"))
-            .then((response) => {
-                if (!response.ok) {
-                    throw new Error(`Failed to load models: ${response.status}`)
-                }
-                return response.json()
-            })
-            .then((data: { models?: unknown }) => {
-                if (cancelled || !Array.isArray(data.models)) {
-                    return
-                }
-
-                const models = data.models.filter(
-                    (model): model is string => typeof model === "string",
-                )
-                if (models.length > 0) {
-                    setDynamicSuggestedModels((current) => ({
-                        ...current,
-                        aihubmix: models,
-                    }))
-                }
-            })
-            .catch((error) => {
-                console.warn("Failed to load AIHubMix models:", error)
-            })
-            .finally(() => {
-                if (cancelled) {
-                    return
-                }
-
-                setLoadedSuggestedProviders((current) => ({
+    const handleFetchModels = async () => {
+        if (!selectedProvider) return
+        const providerId = selectedProvider.id
+        const askedWith = credentialsOf(providerId)
+        setFetchingModels(true)
+        setFetchModelsError("")
+        try {
+            const response = await fetch(
+                getApiEndpoint("/api/provider-models"),
+                {
+                    method: "POST",
+                    headers: {
+                        "Content-Type": "application/json",
+                        "x-access-code":
+                            localStorage.getItem(STORAGE_KEYS.accessCode) || "",
+                    },
+                    body: JSON.stringify({
+                        provider: selectedProvider.provider,
+                        apiKey: selectedProvider.apiKey,
+                        baseUrl: selectedProvider.baseUrl,
+                    }),
+                },
+            )
+            const data = await response.json().catch(() => ({}))
+            if (credentialsOf(providerId) !== askedWith) return
+            // The picker and the error belong to the provider shown
+            const stillShown = selectedProviderIdRef.current === providerId
+            if (Array.isArray(data.models)) {
+                setFetchedModels((current) => ({
                     ...current,
-                    aihubmix: true,
+                    [providerId]: data.models,
                 }))
-                setLoadingSuggestedProvider(null)
-            })
-
-        return () => {
-            cancelled = true
+                if (stillShown) setModelPickerOpen(true)
+            } else if (stillShown) {
+                const hints = dict.errors.llm as Record<string, string>
+                setFetchModelsError(
+                    [hints[data.code], data.error].filter(Boolean).join(" ") ||
+                        `Request failed (${response.status})`,
+                )
+            }
+        } catch {
+            if (
+                selectedProviderIdRef.current === providerId &&
+                credentialsOf(providerId) === askedWith
+            ) {
+                setFetchModelsError(dict.errors.networkError)
+            }
+        } finally {
+            setFetchingModels(false)
         }
-    }, [open, selectedProvider?.provider, loadedSuggestedProviders.aihubmix])
+    }
 
-    // Get suggested models for current provider
-    const suggestedModels = selectedProvider
-        ? dynamicSuggestedModels[selectedProvider.provider] ||
-          SUGGESTED_MODELS[selectedProvider.provider] ||
-          []
+    // The provider's own list once fetched, else the suggested models
+    const suggestedModels: ListedModel[] = selectedProvider
+        ? fetchedModels[selectedProvider.id] ||
+          (SUGGESTED_MODELS[selectedProvider.provider] || []).map((id) => ({
+              id,
+          }))
         : []
-    const isLoadingSuggestedModels =
-        selectedProvider?.provider === loadingSuggestedProvider
+    // Tool calls are what drawing needs: false when known to be missing
+    const supportsTools = (model: ListedModel) =>
+        selectedProvider
+            ? (model.tools ??
+              getModelInfo?.(selectedProvider.provider, model.id)?.tools)
+            : undefined
 
     // Filter out already-added models from suggestions
     const existingModelIds =
         selectedProvider?.models.map((m) => m.modelId) || []
     const availableSuggestions = suggestedModels.filter(
-        (modelId) => !existingModelIds.includes(modelId),
+        (model) => !existingModelIds.includes(model.id),
     )
     const emptyStateSuggestions = selectedProvider
         ? (SUGGESTED_MODELS[selectedProvider.provider] || [])
@@ -246,6 +312,8 @@ export function ModelConfigDialog({
         const newProvider = addProvider(providerType)
         setSelectedProviderId(newProvider.id)
         setValidationStatus("idle")
+        setFetchModelsError("")
+        setModelPickerOpen(false)
     }
 
     // Handle provider field updates
@@ -253,9 +321,9 @@ export function ModelConfigDialog({
         field: keyof ProviderConfig,
         value: string | boolean,
     ) => {
-        if (!selectedProviderId) return
-        updateProvider(selectedProviderId, { [field]: value })
-        // Reset validation when credentials change
+        if (!selectedProviderId || !selectedProvider) return
+        const updates: Partial<ProviderConfig> = { [field]: value }
+        // Reset validation of the provider and its models when credentials change
         const credentialFields = [
             "apiKey",
             "baseUrl",
@@ -266,8 +334,19 @@ export function ModelConfigDialog({
         ]
         if (credentialFields.includes(field)) {
             setValidationStatus("idle")
-            updateProvider(selectedProviderId, { validated: false })
+            setValidatingModelIds(new Set())
+            setFetchedModels(({ [selectedProviderId]: _, ...rest }) => rest)
+            setFetchModelsError("")
+            updates.validated = false
+            updates.models = selectedProvider.models.map((m) => ({
+                ...m,
+                validated: undefined,
+                validationError: undefined,
+                validationWarning: undefined,
+                responseTime: undefined,
+            }))
         }
+        updateProvider(selectedProviderId, updates)
     }
 
     // Handle adding a model to current provider
@@ -337,77 +416,164 @@ export function ModelConfigDialog({
 
         let allValid = true
         let errorCount = 0
+        let idChanged = false
+        const askedWith = credentialsOf(selectedProviderId)
+        const run = ++validationRunRef.current
 
-        // Validate each model
-        for (let i = 0; i < selectedProvider.models.length; i++) {
-            const model = selectedProvider.models[i]
-            setValidatingModelIndex(i)
+        // For EdgeOne, construct baseUrl from current origin
+        const baseUrl = isEdgeOne
+            ? `${window.location.origin}/api/edgeai`
+            : selectedProvider.baseUrl
 
-            try {
-                // For EdgeOne, construct baseUrl from current origin
-                const baseUrl = isEdgeOne
-                    ? `${window.location.origin}/api/edgeai`
-                    : selectedProvider.baseUrl
-
-                const response = await fetch("/api/validate-model", {
-                    method: "POST",
-                    headers: { "Content-Type": "application/json" },
-                    body: JSON.stringify({
-                        provider: selectedProvider.provider,
-                        apiKey: selectedProvider.apiKey,
-                        baseUrl,
-                        modelId: model.modelId,
-                        // AWS Bedrock credentials
-                        awsAccessKeyId: selectedProvider.awsAccessKeyId,
-                        awsSecretAccessKey: selectedProvider.awsSecretAccessKey,
-                        awsRegion: selectedProvider.awsRegion,
-                        // Vertex AI credentials (Express Mode)
-                        vertexApiKey: selectedProvider.vertexApiKey,
-                    }),
-                })
-                const data = await response.json()
-
-                if (data.valid) {
-                    updateModel(selectedProviderId, model.id, {
-                        validated: true,
-                        validationError: undefined,
+        // Test every model at once; each row updates when its answer arrives
+        setValidatingModelIds(new Set(selectedProvider.models.map((m) => m.id)))
+        await Promise.all(
+            selectedProvider.models.map(async (model) => {
+                let update: Partial<ModelConfig>
+                try {
+                    const response = await fetch(
+                        getApiEndpoint("/api/validate-model"),
+                        {
+                            method: "POST",
+                            headers: {
+                                "Content-Type": "application/json",
+                                "x-access-code":
+                                    localStorage.getItem(
+                                        STORAGE_KEYS.accessCode,
+                                    ) || "",
+                            },
+                            body: JSON.stringify({
+                                provider: selectedProvider.provider,
+                                apiKey: selectedProvider.apiKey,
+                                baseUrl,
+                                modelId: model.modelId,
+                                // AWS Bedrock credentials
+                                awsAccessKeyId: selectedProvider.awsAccessKeyId,
+                                awsSecretAccessKey:
+                                    selectedProvider.awsSecretAccessKey,
+                                awsRegion: selectedProvider.awsRegion,
+                                // Temporary AWS credentials, as the chat sends
+                                awsSessionToken:
+                                    selectedProvider.awsSessionToken,
+                                // Vertex AI credentials (Express Mode)
+                                vertexApiKey: selectedProvider.vertexApiKey,
+                            }),
+                        },
+                    )
+                    const data = await response.json().catch(() => ({}))
+                    update = data.valid
+                        ? {
+                              validated: true,
+                              validationError: undefined,
+                              validationWarning: data.warning,
+                              responseTime: data.responseTime,
+                          }
+                        : {
+                              validated: false,
+                              // The hint for the error's kind, then the
+                              // provider's own message
+                              validationError:
+                                  [
+                                      (
+                                          dict.errors.llm as Record<
+                                              string,
+                                              string
+                                          >
+                                      )[data.code],
+                                      data.error,
+                                  ]
+                                      .filter(Boolean)
+                                      .join(" ") ||
+                                  (response.ok
+                                      ? "Validation failed"
+                                      : `Request failed (${response.status})`),
+                              validationWarning: undefined,
+                          }
+                } catch {
+                    update = {
+                        validated: false,
+                        validationError: "Network error",
+                        validationWarning: undefined,
+                    }
+                }
+                // A newer test started: its own results and spinners count,
+                // whatever the credentials are now (they may have come back)
+                if (run !== validationRunRef.current) return
+                // Credentials changed during the test: drop the result. A
+                // change in another tab left the spinner on, so clear it
+                // (model ids are unique, whatever provider is shown).
+                if (credentialsOf(selectedProviderId) !== askedWith) {
+                    setValidatingModelIds((prev) => {
+                        const next = new Set(prev)
+                        next.delete(model.id)
+                        return next
                     })
-                } else {
+                    return
+                }
+                // So did this model's id: the result is for the old one
+                const current = configRef.current.providers
+                    .find((p) => p.id === selectedProviderId)
+                    ?.models.find((m) => m.id === model.id)
+                if (current?.modelId !== model.modelId) {
+                    idChanged = true
+                    setValidatingModelIds((prev) => {
+                        const next = new Set(prev)
+                        next.delete(model.id)
+                        return next
+                    })
+                    return
+                }
+                if (update.validated === false) {
                     allValid = false
                     errorCount++
-                    updateModel(selectedProviderId, model.id, {
-                        validated: false,
-                        validationError: data.error || "Validation failed",
-                    })
                 }
-            } catch {
-                allValid = false
-                errorCount++
-                updateModel(selectedProviderId, model.id, {
-                    validated: false,
-                    validationError: "Network error",
+                updateModel(selectedProviderId, model.id, update)
+                setValidatingModelIds((prev) => {
+                    const next = new Set(prev)
+                    next.delete(model.id)
+                    return next
                 })
+            }),
+        )
+        if (run !== validationRunRef.current) return
+        if (credentialsOf(selectedProviderId) !== askedWith) {
+            // The status line is about the provider shown now
+            if (selectedProviderIdRef.current === selectedProviderId) {
+                setValidationStatus("idle")
             }
+            return
         }
 
-        setValidatingModelIndex(null)
-
-        if (allValid) {
-            setValidationStatus("success")
+        // A model whose id changed was not tested
+        if (allValid && !idChanged) {
             updateProvider(selectedProviderId, { validated: true })
+        }
+        // The status line is about the provider shown now
+        if (selectedProviderIdRef.current !== selectedProviderId) return
+        if (idChanged) {
+            setValidationStatus("idle")
+        } else if (allValid) {
+            setValidationStatus("success")
             // Reset to idle after showing success briefly (with cleanup)
             if (validationResetTimeoutRef.current) {
                 clearTimeout(validationResetTimeoutRef.current)
             }
             validationResetTimeoutRef.current = setTimeout(() => {
-                setValidationStatus("idle")
                 validationResetTimeoutRef.current = null
+                if (run !== validationRunRef.current) return
+                setValidationStatus("idle")
             }, 1500)
         } else {
             setValidationStatus("error")
             setValidationError(`${errorCount} model(s) failed validation`)
         }
-    }, [selectedProvider, selectedProviderId, updateProvider, updateModel])
+    }, [
+        selectedProvider,
+        selectedProviderId,
+        updateProvider,
+        updateModel,
+        dict,
+    ])
 
     // Get all available provider types
     const availableProviders = Object.keys(PROVIDER_INFO) as ProviderName[]
@@ -563,6 +729,10 @@ export function ModelConfigDialog({
                                                 )
                                                 setValidationStatus("idle")
                                                 setShowApiKey(false)
+                                                // These belong to the
+                                                // provider shown before
+                                                setFetchModelsError("")
+                                                setModelPickerOpen(false)
                                             }}
                                             className={cn(
                                                 "group flex items-center gap-3 px-3 py-2.5 rounded-xl w-full",
@@ -615,7 +785,9 @@ export function ModelConfigDialog({
 
                         {/* Add Provider */}
                         <div className="p-3 border-t border-border-subtle">
+                            {/* Always empty so picking the same type again still fires */}
                             <Select
+                                value=""
                                 onValueChange={(v) =>
                                     handleAddProvider(v as ProviderName)
                                 }
@@ -836,57 +1008,136 @@ export function ModelConfigDialog({
                                                 >
                                                     <Plus className="h-3.5 w-3.5" />
                                                 </Button>
-                                                <Select
-                                                    onValueChange={(value) => {
-                                                        if (value) {
-                                                            handleAddModel(
-                                                                value,
-                                                            )
+                                                {PROVIDER_INFO[
+                                                    selectedProvider.provider
+                                                ].modelList && (
+                                                    <Button
+                                                        variant="outline"
+                                                        size="sm"
+                                                        className="h-8 rounded-lg"
+                                                        onClick={
+                                                            handleFetchModels
                                                         }
-                                                    }}
-                                                    disabled={
-                                                        isLoadingSuggestedModels ||
-                                                        availableSuggestions.length ===
-                                                            0
-                                                    }
-                                                >
-                                                    <SelectTrigger className="w-28 h-8 rounded-lg hover:bg-interactive-hover">
-                                                        {isLoadingSuggestedModels ? (
+                                                        disabled={
+                                                            fetchingModels
+                                                        }
+                                                        title={
+                                                            dict.modelConfig
+                                                                .fetchModels
+                                                        }
+                                                        aria-label={
+                                                            dict.modelConfig
+                                                                .fetchModels
+                                                        }
+                                                    >
+                                                        {fetchingModels ? (
                                                             <Loader2 className="h-3.5 w-3.5 animate-spin" />
                                                         ) : (
-                                                            <span className="text-xs">
-                                                                {availableSuggestions.length ===
+                                                            <RefreshCw className="h-3.5 w-3.5" />
+                                                        )}
+                                                    </Button>
+                                                )}
+                                                {/* modal: the dialog blocks the
+                                                wheel outside itself, and the
+                                                list is rendered outside it */}
+                                                <Popover
+                                                    modal
+                                                    open={modelPickerOpen}
+                                                    onOpenChange={
+                                                        setModelPickerOpen
+                                                    }
+                                                >
+                                                    <PopoverTrigger asChild>
+                                                        <Button
+                                                            variant="outline"
+                                                            size="sm"
+                                                            className="w-28 h-8 rounded-lg text-xs"
+                                                            disabled={
+                                                                availableSuggestions.length ===
                                                                 0
-                                                                    ? dict
-                                                                          .modelConfig
-                                                                          .allAdded
-                                                                    : dict
-                                                                          .modelConfig
-                                                                          .suggested}
-                                                            </span>
-                                                        )}
-                                                    </SelectTrigger>
-                                                    <SelectContent className="max-h-72">
-                                                        {availableSuggestions.map(
-                                                            (modelId) => (
-                                                                <SelectItem
-                                                                    key={
-                                                                        modelId
+                                                            }
+                                                        >
+                                                            {availableSuggestions.length ===
+                                                            0
+                                                                ? dict
+                                                                      .modelConfig
+                                                                      .allAdded
+                                                                : dict
+                                                                      .modelConfig
+                                                                      .suggested}
+                                                        </Button>
+                                                    </PopoverTrigger>
+                                                    <PopoverContent
+                                                        className="w-80 p-0"
+                                                        align="end"
+                                                    >
+                                                        <Command>
+                                                            <CommandInput
+                                                                placeholder={
+                                                                    dict
+                                                                        .modelConfig
+                                                                        .searchModels
+                                                                }
+                                                            />
+                                                            <CommandList className="max-h-72">
+                                                                <CommandEmpty>
+                                                                    {
+                                                                        dict
+                                                                            .modelConfig
+                                                                            .noModelsFound
                                                                     }
-                                                                    value={
-                                                                        modelId
-                                                                    }
-                                                                    className="font-mono text-xs"
-                                                                >
-                                                                    {modelId}
-                                                                </SelectItem>
-                                                            ),
-                                                        )}
-                                                    </SelectContent>
-                                                </Select>
+                                                                </CommandEmpty>
+                                                                {availableSuggestions.map(
+                                                                    (model) => (
+                                                                        <CommandItem
+                                                                            key={
+                                                                                model.id
+                                                                            }
+                                                                            value={
+                                                                                model.id
+                                                                            }
+                                                                            onSelect={() => {
+                                                                                handleAddModel(
+                                                                                    model.id,
+                                                                                )
+                                                                                setModelPickerOpen(
+                                                                                    false,
+                                                                                )
+                                                                            }}
+                                                                            className="font-mono text-xs"
+                                                                        >
+                                                                            <span className="truncate">
+                                                                                {
+                                                                                    model.id
+                                                                                }
+                                                                            </span>
+                                                                            {supportsTools(
+                                                                                model,
+                                                                            ) ===
+                                                                                false && (
+                                                                                <span className="ml-auto shrink-0 font-sans text-[10px] text-amber-600 dark:text-amber-400">
+                                                                                    {
+                                                                                        dict
+                                                                                            .modelConfig
+                                                                                            .noTools
+                                                                                    }
+                                                                                </span>
+                                                                            )}
+                                                                        </CommandItem>
+                                                                    ),
+                                                                )}
+                                                            </CommandList>
+                                                        </Command>
+                                                    </PopoverContent>
+                                                </Popover>
                                             </div>
                                         }
                                     >
+                                        {fetchModelsError && (
+                                            <p className="mb-2 text-xs text-destructive">
+                                                {fetchModelsError}
+                                            </p>
+                                        )}
                                         {/* Model List */}
                                         <div className="rounded-2xl border border-border-subtle bg-surface-2/30 overflow-hidden min-h-[120px]">
                                             {selectedProvider.models.length ===
@@ -940,7 +1191,7 @@ export function ModelConfigDialog({
                                             ) : (
                                                 <div className="divide-y divide-border-subtle">
                                                     {selectedProvider.models.map(
-                                                        (model, index) => (
+                                                        (model) => (
                                                             <div
                                                                 key={model.id}
                                                                 className={cn(
@@ -950,28 +1201,24 @@ export function ModelConfigDialog({
                                                                 <div className="flex items-center gap-3 p-3 min-w-0">
                                                                     {/* Status icon */}
                                                                     <div className="flex items-center justify-center w-8 h-8 rounded-lg flex-shrink-0">
-                                                                        {validatingModelIndex !==
-                                                                            null &&
-                                                                        index ===
-                                                                            validatingModelIndex ? (
+                                                                        {validatingModelIds.has(
+                                                                            model.id,
+                                                                        ) ? (
                                                                             // Currently validating
                                                                             <div className="w-full h-full rounded-lg bg-blue-500/10 flex items-center justify-center">
                                                                                 <Loader2 className="h-4 w-4 text-blue-500 animate-spin" />
                                                                             </div>
-                                                                        ) : validatingModelIndex !==
-                                                                              null &&
-                                                                          index >
-                                                                              validatingModelIndex &&
-                                                                          model.validated ===
-                                                                              undefined ? (
-                                                                            // Queued
-                                                                            <div className="w-full h-full rounded-lg bg-muted flex items-center justify-center">
-                                                                                <Clock className="h-4 w-4 text-muted-foreground" />
-                                                                            </div>
                                                                         ) : model.validated ===
                                                                           true ? (
-                                                                            // Valid
-                                                                            <div className="w-full h-full rounded-lg bg-success-muted flex items-center justify-center">
+                                                                            // Valid, with the time the test took
+                                                                            <div
+                                                                                className="w-full h-full rounded-lg bg-success-muted flex items-center justify-center"
+                                                                                title={
+                                                                                    model.responseTime
+                                                                                        ? `${(model.responseTime / 1000).toFixed(1)} s`
+                                                                                        : undefined
+                                                                                }
+                                                                            >
                                                                                 <Check className="h-4 w-4 text-success" />
                                                                             </div>
                                                                         ) : model.validated ===
@@ -989,7 +1236,10 @@ export function ModelConfigDialog({
                                                                     </div>
                                                                     <Input
                                                                         value={
-                                                                            model.modelId
+                                                                            modelIdDraft?.id ===
+                                                                            model.id
+                                                                                ? modelIdDraft.value
+                                                                                : model.modelId
                                                                         }
                                                                         title={
                                                                             model.modelId
@@ -1007,24 +1257,14 @@ export function ModelConfigDialog({
                                                                                     null,
                                                                                 )
                                                                             }
-                                                                            if (
-                                                                                selectedProviderId
-                                                                            ) {
-                                                                                updateModel(
-                                                                                    selectedProviderId,
-                                                                                    model.id,
-                                                                                    {
-                                                                                        modelId:
-                                                                                            e
-                                                                                                .target
-                                                                                                .value,
-                                                                                        validated:
-                                                                                            undefined,
-                                                                                        validationError:
-                                                                                            undefined,
-                                                                                    },
-                                                                                )
-                                                                            }
+                                                                            setModelIdDraft(
+                                                                                {
+                                                                                    id: model.id,
+                                                                                    value: e
+                                                                                        .target
+                                                                                        .value,
+                                                                                },
+                                                                            )
                                                                         }}
                                                                         onKeyDown={(
                                                                             e,
@@ -1041,6 +1281,10 @@ export function ModelConfigDialog({
                                                                         ) => {
                                                                             const newModelId =
                                                                                 e.target.value.trim()
+                                                                            // Drop the draft; an invalid ID falls back to the saved one
+                                                                            setModelIdDraft(
+                                                                                null,
+                                                                            )
 
                                                                             // Helper to show error with shake
                                                                             const showError =
@@ -1135,6 +1379,28 @@ export function ModelConfigDialog({
                                                                             setEditError(
                                                                                 null,
                                                                             )
+                                                                            if (
+                                                                                selectedProviderId &&
+                                                                                newModelId !==
+                                                                                    model.modelId
+                                                                            ) {
+                                                                                updateModel(
+                                                                                    selectedProviderId,
+                                                                                    model.id,
+                                                                                    {
+                                                                                        modelId:
+                                                                                            newModelId,
+                                                                                        validated:
+                                                                                            undefined,
+                                                                                        validationError:
+                                                                                            undefined,
+                                                                                        validationWarning:
+                                                                                            undefined,
+                                                                                        responseTime:
+                                                                                            undefined,
+                                                                                    },
+                                                                                )
+                                                                            }
                                                                         }}
                                                                         className="flex-1 min-w-0 font-mono text-sm h-8 border-0 bg-transparent focus-visible:bg-background focus-visible:ring-1"
                                                                     />
@@ -1162,6 +1428,28 @@ export function ModelConfigDialog({
                                                                             }
                                                                         </p>
                                                                     )}
+                                                                {!model.validationWarning &&
+                                                                    getModelInfo?.(
+                                                                        selectedProvider.provider,
+                                                                        model.modelId,
+                                                                    )?.tools ===
+                                                                        false && (
+                                                                        <p className="text-[11px] text-amber-600 dark:text-amber-400 px-3 pb-2 pl-14">
+                                                                            {
+                                                                                dict
+                                                                                    .modelConfig
+                                                                                    .mayNotDraw
+                                                                            }
+                                                                        </p>
+                                                                    )}
+                                                                {model.validated &&
+                                                                    model.validationWarning && (
+                                                                        <p className="text-[11px] text-amber-600 dark:text-amber-400 px-3 pb-2 pl-14">
+                                                                            {
+                                                                                model.validationWarning
+                                                                            }
+                                                                        </p>
+                                                                    )}
                                                                 {/* Show edit error inline */}
                                                                 {editError?.modelId ===
                                                                     model.id && (

components/model-selector.tsx

@@ -264,9 +264,13 @@ export function ModelSelector({
                                                         (model) => (
                                                             <ModelSelectorItem
                                                                 key={model.id}
-                                                                value={
-                                                                    model.modelId
-                                                                }
+                                                                // Unique value so same-named models highlight
+                                                                // separately; keywords keep search by name
+                                                                value={model.id}
+                                                                keywords={[
+                                                                    model.modelId,
+                                                                    providerLabel,
+                                                                ]}
                                                                 onSelect={() =>
                                                                     handleSelect(
                                                                         model.id,
@@ -351,9 +355,11 @@ export function ModelSelector({
                                                         (model) => (
                                                             <ModelSelectorItem
                                                                 key={model.id}
-                                                                value={
-                                                                    model.modelId
-                                                                }
+                                                                value={model.id}
+                                                                keywords={[
+                                                                    model.modelId,
+                                                                    providerLabel,
+                                                                ]}
                                                                 onSelect={() =>
                                                                     handleSelect(
                                                                         model.id,

components/provider-credentials-fields.tsx

@@ -13,7 +13,12 @@ import {
 } from "@/components/ui/select"
 import { useDictionary } from "@/hooks/use-dictionary"
 import { formatMessage } from "@/lib/i18n/utils"
-import { PROVIDER_INFO, type ProviderName } from "@/lib/types/model-config"
+import {
+    chatRequestUrl,
+    normalizeBaseUrl,
+    PROVIDER_INFO,
+    type ProviderName,
+} from "@/lib/types/model-config"
 
 // Logical secret field. The caller owns the actual input — plaintext for the
 // user dialog, write-only masked for the admin panel — supplied via
@@ -26,7 +31,7 @@ export type SecretField =
     | "vertexApiKey"
 
 // AWS regions offered for Bedrock (shared by both screens)
-const AWS_REGIONS: Array<[string, string]> = [
+export const AWS_REGIONS: Array<[string, string]> = [
     ["us-east-1", "N. Virginia"],
     ["us-east-2", "Ohio"],
     ["us-west-2", "Oregon"],
@@ -77,6 +82,7 @@ export function ProviderCredentialsFields({
     const baseUrlLabel = formatMessage(dict.modelConfig.baseUrlWithExample, {
         example: info.defaultBaseUrl || "https://api.example.com/v1",
     })
+    const requestUrl = baseUrl ? chatRequestUrl(provider, baseUrl) : null
 
     // EdgeOne needs no credentials — the caller supplies just a test button
     if (provider === "edgeone") {
@@ -208,17 +214,29 @@ export function ProviderCredentialsFields({
                 </>
             ) : (
                 <>
-                    {/* API Key */}
+                    {/* API Key, with a link to where the provider issues keys */}
                     <div className="space-y-2">
-                        <Label
-                            htmlFor="api-key"
-                            className="text-xs font-medium flex items-center gap-1.5"
-                        >
-                            <Key className="h-3.5 w-3.5 text-muted-foreground" />
-                            {dict.modelConfig.apiKey}
-                            {provider === "ollama" &&
-                                ` ${dict.modelConfig.optional}`}
-                        </Label>
+                        <div className="flex items-center justify-between">
+                            <Label
+                                htmlFor="api-key"
+                                className="text-xs font-medium flex items-center gap-1.5"
+                            >
+                                <Key className="h-3.5 w-3.5 text-muted-foreground" />
+                                {dict.modelConfig.apiKey}
+                                {provider === "ollama" &&
+                                    ` ${dict.modelConfig.optional}`}
+                            </Label>
+                            {info.apiKeyUrl && (
+                                <a
+                                    href={info.apiKeyUrl}
+                                    target="_blank"
+                                    rel="noopener noreferrer"
+                                    className="text-xs text-primary hover:underline"
+                                >
+                                    {dict.modelConfig.getApiKey}
+                                </a>
+                            )}
+                        </div>
                         {renderSecret({ field: "apiKey", id: "api-key" })}
                     </div>
 
@@ -238,12 +256,28 @@ export function ProviderCredentialsFields({
                             onChange={(e) =>
                                 onChange("baseUrl", e.target.value)
                             }
+                            // Drop a pasted endpoint path such as /chat/completions
+                            onBlur={(e) => {
+                                const normalized = normalizeBaseUrl(
+                                    e.target.value,
+                                )
+                                if (normalized !== e.target.value) {
+                                    onChange("baseUrl", normalized)
+                                }
+                            }}
                             placeholder={
                                 info.defaultBaseUrl ||
                                 dict.modelConfig.customEndpoint
                             }
                             className="h-9 rounded-xl font-mono text-xs"
                         />
+                        {requestUrl && (
+                            <p className="text-xs text-muted-foreground font-mono break-all">
+                                {formatMessage(dict.modelConfig.requestUrl, {
+                                    url: requestUrl,
+                                })}
+                            </p>
+                        )}
                         {provider === "minimax" && (
                             <p className="text-xs text-muted-foreground">
                                 {dict.modelConfig.minimaxBaseUrlHint}

components/settings-dialog.tsx

@@ -75,6 +75,8 @@ interface SettingsDialogProps {
     onOpenModelConfig?: () => void
     customSystemMessage?: string
     onCustomSystemMessageChange?: (value: string) => void
+    maxOutputTokens?: string
+    onMaxOutputTokensChange?: (value: string) => void
 }
 
 export const STORAGE_ACCESS_CODE_KEY = "next-ai-draw-io-access-code"
@@ -101,6 +103,8 @@ function SettingsContent({
     onOpenModelConfig,
     customSystemMessage = "",
     onCustomSystemMessageChange = () => {},
+    maxOutputTokens = "",
+    onMaxOutputTokensChange = () => {},
 }: SettingsDialogProps) {
     const dict = useDictionary()
     const router = useRouter()
@@ -591,6 +595,24 @@ function SettingsContent({
                         />
                     </div>
 
+                    {/* Max Output Tokens */}
+                    <SettingItem
+                        label={dict.settings.maxOutputTokens}
+                        description={dict.settings.maxOutputTokensDescription}
+                    >
+                        <Input
+                            id="max-output-tokens"
+                            type="text"
+                            inputMode="numeric"
+                            value={maxOutputTokens}
+                            onChange={(e) =>
+                                onMaxOutputTokensChange(e.target.value)
+                            }
+                            placeholder="64000"
+                            className="h-9 w-28 text-sm"
+                        />
+                    </SettingItem>
+
                     {/* Send Shortcut */}
                     <SettingItem
                         label={dict.settings.sendShortcut}

contexts/diagram-context.tsx

@@ -1,26 +1,34 @@
 "use client"
 
 import type React from "react"
-import { createContext, useContext, useEffect, useRef, useState } from "react"
+import { createContext, useCallback, useContext, useRef, useState } from "react"
 import type { DrawIoEmbedRef, EventExport } from "react-drawio"
 import { toast } from "sonner"
 import type { ExportFormat } from "@/components/save-dialog"
 import { getApiEndpoint } from "@/lib/base-path"
 import {
-    extractDiagramXML,
-    isRealDiagram,
-    validateAndFixXml,
-} from "../lib/utils"
+    BLANK_MXFILE,
+    normalizeToMxfile,
+} from "@/packages/mcp-server/src/pages.ts"
+import { validateAndFixXml } from "@/packages/mcp-server/src/xml-validation.ts"
+import { extractDiagramXML, isRealDiagram } from "../lib/utils"
 
 interface DiagramContextType {
     chartXML: string
+    // chartXML right away, before the re-render (loadDiagram sets both)
+    chartXMLRef: React.MutableRefObject<string>
     latestSvg: string
     diagramHistory: { svg: string; xml: string }[]
     setDiagramHistory: (history: { svg: string; xml: string }[]) => void
     loadDiagram: (chart: string, skipValidation?: boolean) => string | null
-    handleExport: () => void
-    handleExportWithoutHistory: () => void
-    resolverRef: React.MutableRefObject<((value: string) => void) | null>
+    // Both return the export's tag (empty when draw.io is not there yet)
+    handleExport: () => string
+    handleExportWithoutHistory: () => string
+    // Pending exports by tag; a history or plain export's resolver gets the
+    // first page's XML
+    exportResolversRef: React.MutableRefObject<
+        Record<string, (data: string, xml?: string) => void>
+    >
     drawioRef: React.MutableRefObject<DrawIoEmbedRef | null>
     handleDiagramExport: (data: EventExport) => void
     handleDiagramAutoSave: (data: { xml?: string }) => void
@@ -42,6 +50,12 @@ interface DiagramContextType {
 
 const DiagramContext = createContext<DiagramContextType | undefined>(undefined)
 
+// Every export carries a tag in the request's `message` field. draw.io
+// echoes the request back in the export event, so each result reaches its
+// own caller. Tags end in a request number, so a late result never answers
+// a newer request.
+type ExportTag = "thumbnail" | "validation"
+
 export function DiagramProvider({ children }: { children: React.ReactNode }) {
     const [chartXML, setChartXML] = useState<string>("")
     const [latestSvg, setLatestSvg] = useState<string>("")
@@ -52,11 +66,13 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
     const [showSaveDialog, setShowSaveDialog] = useState(false)
     const hasCalledOnLoadRef = useRef(false)
     const drawioRef = useRef<DrawIoEmbedRef | null>(null)
-    const resolverRef = useRef<((value: string) => void) | null>(null)
-    // Resolver for PNG export (used for VLM validation)
-    const pngResolverRef = useRef<((value: string) => void) | null>(null)
-    // Track if we're expecting an export for history (user-initiated)
-    const expectHistoryExportRef = useRef<boolean>(false)
+    // Pending exports, keyed by their export tag
+    const exportResolversRef = useRef<
+        Record<string, (data: string, xml?: string) => void>
+    >({})
+    // Pending history exports: the document each one was asked for
+    const historyXmlRef = useRef(new Map<string, string>())
+    const exportSeqRef = useRef(0)
     // Track latest chartXML for restoration after remount
     const chartXMLRef = useRef<string>("")
 
@@ -76,94 +92,85 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         setIsDrawioReady(false)
     }
 
-    // Keep chartXMLRef in sync with state for restoration after remount
-    useEffect(() => {
-        chartXMLRef.current = chartXML
-    }, [chartXML])
-
-    // Track if we're expecting an export for file save (stores raw export data)
-    const saveResolverRef = useRef<{
-        resolver: ((data: string, fullDiagramXML?: string) => void) | null
-        format: ExportFormat | null
-    }>({ resolver: null, format: null })
+    // Update chartXML and its ref together, so callbacks that read the ref
+    // (export handler, autosave) see the new value right away
+    const updateChartXML = (xml: string) => {
+        chartXMLRef.current = xml
+        setChartXML(xml)
+    }
 
     const handleExport = () => {
-        if (drawioRef.current) {
-            // Mark that this export should be saved to history
-            expectHistoryExportRef.current = true
-            drawioRef.current.exportDiagram({
-                format: "xmlsvg",
-            })
-        }
+        if (!drawioRef.current) return ""
+        // Save this export to history, with the document shown now:
+        // chartXML can change before the result comes back
+        const tag = `history-${++exportSeqRef.current}`
+        historyXmlRef.current.set(tag, chartXMLRef.current)
+        drawioRef.current.exportDiagram({
+            format: "xmlsvg",
+            message: tag,
+        })
+        return tag
     }
 
     const handleExportWithoutHistory = () => {
-        if (drawioRef.current) {
-            // Export without saving to history (for edit_diagram fetching current state)
-            drawioRef.current.exportDiagram({
-                format: "xmlsvg",
-            })
-        }
+        if (!drawioRef.current) return ""
+        // Export without saving to history (for edit_diagram fetching current state)
+        const tag = `fetch-${++exportSeqRef.current}`
+        drawioRef.current.exportDiagram({
+            format: "xmlsvg",
+            message: tag,
+        })
+        return tag
     }
 
-    // Get current diagram as SVG for thumbnail (used by session storage)
-    const getThumbnailSvg = async (): Promise<string | null> => {
+    // Export with a tag in `message` (draw.io echoes it back in the export
+    // event) and wait for that result. Resolves to null on timeout, which is
+    // expected occasionally.
+    // (Reads refs only, so it keeps one identity)
+    const requestTaggedExport = useCallback(
+        (tag: ExportTag, format: "xmlsvg" | "png", timeoutMs: number) =>
+            new Promise<string | null>((resolve) => {
+                const id = `${tag}-${++exportSeqRef.current}`
+                const finish = (value: string | null) => {
+                    clearTimeout(timer)
+                    delete exportResolversRef.current[id]
+                    resolve(value)
+                }
+                const timer = setTimeout(() => finish(null), timeoutMs)
+                exportResolversRef.current[id] = finish
+                drawioRef.current?.exportDiagram({ format, message: id })
+            }),
+        [],
+    )
+
+    // Get current diagram as SVG for thumbnail (used by session storage).
+    // One identity: the chat's auto-save depends on it, and each thumbnail
+    // renders this provider again (latestSvg), which would otherwise start
+    // the next save
+    const getThumbnailSvg = useCallback(async (): Promise<string | null> => {
         if (!drawioRef.current) return null
         // Don't export if diagram is empty
-        if (!isRealDiagram(chartXML)) return null
-
-        try {
-            const svgData = await Promise.race([
-                new Promise<string>((resolve) => {
-                    resolverRef.current = resolve
-                    drawioRef.current?.exportDiagram({ format: "xmlsvg" })
-                }),
-                new Promise<string>((_, reject) =>
-                    setTimeout(() => reject(new Error("Export timeout")), 3000),
-                ),
-            ])
+        if (!isRealDiagram(chartXMLRef.current)) return null
 
+        // xmlsvg exports return an SVG data URL
+        const svgData = await requestTaggedExport("thumbnail", "xmlsvg", 3000)
+        if (svgData?.startsWith("data:image/svg")) {
             // Update latestSvg so it's available for future saves
-            if (svgData?.includes("<svg")) {
-                setLatestSvg(svgData)
-                return svgData
-            }
-            return null
-        } catch {
-            // Timeout is expected occasionally - don't log as error
-            return null
+            setLatestSvg(svgData)
+            return svgData
         }
-    }
+        return null
+    }, [requestTaggedExport])
 
     // Capture current diagram as PNG for VLM validation
     const captureValidationPng = async (): Promise<string | null> => {
         if (!drawioRef.current) return null
         // Don't export if diagram is empty
-        if (!isRealDiagram(chartXML)) return null
+        if (!isRealDiagram(chartXMLRef.current)) return null
 
-        try {
-            const pngData = await Promise.race([
-                new Promise<string>((resolve) => {
-                    pngResolverRef.current = resolve
-                    drawioRef.current?.exportDiagram({ format: "png" })
-                }),
-                new Promise<string>((_, reject) =>
-                    setTimeout(
-                        () => reject(new Error("PNG export timeout")),
-                        5000,
-                    ),
-                ),
-            ])
-
-            // PNG data should be a base64 data URL
-            if (pngData?.startsWith("data:image/png")) {
-                return pngData
-            }
-            return null
-        } catch {
-            // Timeout is expected occasionally - don't log as error
-            return null
-        }
+        const pngData = await requestTaggedExport("validation", "png", 5000)
+        // PNG data should be a base64 data URL
+        return pngData?.startsWith("data:image/png") ? pngData : null
     }
 
     const loadDiagram = (
@@ -172,9 +179,11 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
     ): string | null => {
         let xmlToLoad = chart
 
-        // Validate XML structure before loading (unless skipped for internal use)
+        // Validate XML structure before loading (unless skipped for internal
+        // use). Not strict: the XML may hold the user's own diagram, and the
+        // tool handlers check model XML strictly before it gets here.
         if (!skipValidation) {
-            const validation = validateAndFixXml(chart)
+            const validation = validateAndFixXml(chart, { strict: false })
             if (!validation.valid) {
                 console.warn(
                     "[loadDiagram] Validation error:",
@@ -193,7 +202,7 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         }
 
         // Keep chartXML in sync even when diagrams are injected (e.g., display_diagram tool)
-        setChartXML(xmlToLoad)
+        updateChartXML(xmlToLoad)
 
         if (drawioRef.current) {
             drawioRef.current.load({
@@ -205,26 +214,14 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
     }
 
     const handleDiagramExport = (data: EventExport) => {
-        // Handle PNG export for VLM validation
-        if (pngResolverRef.current && data.data?.startsWith("data:image/png")) {
-            pngResolverRef.current(data.data)
-            pngResolverRef.current = null
+        // Thumbnail, validation PNG and file save exports go only to their
+        // own caller
+        const tag = data.message?.message
+        if (/^(thumbnail|validation|save)-/.test(tag ?? "")) {
+            exportResolversRef.current[tag as string]?.(data.data, data.xml)
             return
         }
 
-        // Handle save to file if requested (process raw data before extraction)
-        if (saveResolverRef.current.resolver) {
-            const format = saveResolverRef.current.format
-            saveResolverRef.current.resolver(data.data, data.xml)
-            saveResolverRef.current = { resolver: null, format: null }
-            // For non-xmlsvg formats, skip XML extraction as it will fail
-            // Only drawio (which uses xmlsvg internally) has the content attribute
-            // xmlsvg is saved directly as SVG file, no need for extraction
-            if (format === "png" || format === "svg" || format === "xmlsvg") {
-                return
-            }
-        }
-
         // Don't write chartXML here: exports don't change the diagram, and
         // data.xml from xmlsvg exports has compressed <diagram> payloads that
         // would break edit_diagram/display_diagram. Autosave keeps chartXML
@@ -235,39 +232,49 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         // Only add to history if this was a user-initiated export
         // Limit to 20 entries to prevent memory leaks during long sessions
         const MAX_HISTORY_SIZE = 20
-        if (expectHistoryExportRef.current) {
+        const askedXml =
+            tag !== undefined ? historyXmlRef.current.get(tag) : undefined
+        if (askedXml !== undefined) {
+            historyXmlRef.current.delete(tag as string)
+            // Store the full multi-page document (extractedXML is only the
+            // first page), so restoring a version keeps every page
+            const historyXml = askedXml || extractedXML
             setDiagramHistory((prev) => {
                 const newHistory = [
                     ...prev,
                     {
                         svg: data.data,
-                        xml: extractedXML,
+                        xml: historyXml,
                     },
                 ]
                 // Keep only the last MAX_HISTORY_SIZE entries (circular buffer)
                 return newHistory.slice(-MAX_HISTORY_SIZE)
             })
-            expectHistoryExportRef.current = false
         }
 
-        if (resolverRef.current) {
-            resolverRef.current(extractedXML)
-            resolverRef.current = null
+        // The chat's own export (onFetchChart), not another one in flight
+        const resolve =
+            tag !== undefined ? exportResolversRef.current[tag] : undefined
+        if (resolve) {
+            delete exportResolversRef.current[tag as string]
+            resolve(extractedXML)
         }
     }
 
+    // react-drawio registers this callback once per iframe mount, so it must
+    // read refs: state captured in its closure would stay stale after a remount
     const handleDiagramAutoSave = (data: { xml?: string }) => {
         if (!data?.xml) return
-        // Don't overwrite a pending restore - if we have a real diagram in state
-        // but DrawIO isn't ready yet, it means we're waiting to restore
-        if (!isDrawioReady && isRealDiagram(chartXML)) {
+        // Don't overwrite a pending restore - if we have a real diagram but
+        // DrawIO hasn't loaded yet, it means we're waiting to restore
+        if (!hasCalledOnLoadRef.current && isRealDiagram(chartXMLRef.current)) {
             return
         }
-        setChartXML(data.xml)
+        updateChartXML(data.xml)
     }
 
     const clearDiagram = () => {
-        const emptyDiagram = `<mxfile><diagram name="Page-1" id="page-1"><mxGraphModel><root><mxCell id="0"/><mxCell id="1" parent="0"/></root></mxGraphModel></diagram></mxfile>`
+        const emptyDiagram = BLANK_MXFILE
         // Skip validation for trusted internal template (loadDiagram also sets chartXML)
         loadDiagram(emptyDiagram, true)
         setLatestSvg("")
@@ -289,83 +296,88 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         const drawioFormat =
             format === "drawio" || format === "xmlsvg" ? "xmlsvg" : format
 
-        // Set up the resolver before triggering export
-        saveResolverRef.current = {
-            resolver: (exportData: string, fullDiagramXML?: string) => {
-                let fileContent: string | Blob
-                let mimeType: string
-                let extension: string
+        // Each save has its own tag, so two at once never swap results
+        const tag = `save-${++exportSeqRef.current}`
+        exportResolversRef.current[tag] = (
+            exportData: string,
+            fullDiagramXML?: string,
+        ) => {
+            delete exportResolversRef.current[tag]
+            let fileContent: string | Blob
+            let mimeType: string
+            let extension: string
 
-                if (format === "drawio") {
-                    // Prefer the complete document from the export event so all pages are saved.
-                    const xml = fullDiagramXML?.trim()
-                        ? fullDiagramXML
-                        : extractDiagramXML(exportData)
-                    let xmlContent = xml
-                    if (!xml.includes("<mxfile")) {
-                        xmlContent = `<mxfile><diagram name="Page-1" id="page-1">${xml}</diagram></mxfile>`
-                    }
-                    fileContent = xmlContent
-                    mimeType = "application/xml"
-                    extension = ".drawio"
-                } else if (format === "png") {
-                    // PNG data comes as base64 data URL
-                    fileContent = exportData
-                    mimeType = "image/png"
-                    extension = ".png"
-                } else if (format === "xmlsvg") {
-                    // Editable SVG: pass data URL directly (like PNG)
-                    fileContent = exportData
-                    mimeType = "image/svg+xml"
-                    extension = ".drawio.svg"
-                } else {
-                    // SVG format (view-only)
-                    fileContent = exportData
-                    mimeType = "image/svg+xml"
-                    extension = ".svg"
-                }
+            if (format === "drawio") {
+                // Prefer the complete document from the export event so all pages are saved.
+                const xml = fullDiagramXML?.trim()
+                    ? fullDiagramXML
+                    : extractDiagramXML(exportData)
+                fileContent =
+                    normalizeToMxfile(xml, {
+                        pageId: "page-1",
+                        pageName: "Page-1",
+                    }) ?? xml
+                mimeType = "application/xml"
+                extension = ".drawio"
+            } else if (format === "png") {
+                // PNG data comes as base64 data URL
+                fileContent = exportData
+                mimeType = "image/png"
+                extension = ".png"
+            } else if (format === "xmlsvg") {
+                // Editable SVG: pass data URL directly (like PNG)
+                fileContent = exportData
+                mimeType = "image/svg+xml"
+                extension = ".drawio.svg"
+            } else {
+                // SVG format (view-only)
+                fileContent = exportData
+                mimeType = "image/svg+xml"
+                extension = ".svg"
+            }
 
-                // Log save event to Langfuse (flags the trace)
-                logSaveToLangfuse(filename, format, sessionId)
+            // Log save event to Langfuse (flags the trace)
+            logSaveToLangfuse(filename, format, sessionId)
 
-                // Handle download
-                let url: string
-                if (
-                    typeof fileContent === "string" &&
-                    fileContent.startsWith("data:")
-                ) {
-                    // Already a data URL (PNG)
-                    url = fileContent
-                } else {
-                    const blob = new Blob([fileContent], { type: mimeType })
-                    url = URL.createObjectURL(blob)
-                }
+            // Handle download
+            let url: string
+            if (
+                typeof fileContent === "string" &&
+                fileContent.startsWith("data:")
+            ) {
+                // Already a data URL (PNG)
+                url = fileContent
+            } else {
+                const blob = new Blob([fileContent], { type: mimeType })
+                url = URL.createObjectURL(blob)
+            }
 
-                const a = document.createElement("a")
-                a.href = url
-                a.download = `${filename}${extension}`
-                document.body.appendChild(a)
-                a.click()
-                document.body.removeChild(a)
+            const a = document.createElement("a")
+            a.href = url
+            a.download = `${filename}${extension}`
+            document.body.appendChild(a)
+            a.click()
+            document.body.removeChild(a)
 
-                // Show success toast after download is initiated
-                if (successMessage) {
-                    toast.success(successMessage, {
-                        position: "bottom-left",
-                        duration: 2500,
-                    })
-                }
+            // Show success toast after download is initiated
+            if (successMessage) {
+                toast.success(successMessage, {
+                    position: "bottom-left",
+                    duration: 2500,
+                })
+            }
 
-                // Delay URL revocation to ensure download completes
-                if (!url.startsWith("data:")) {
-                    setTimeout(() => URL.revokeObjectURL(url), 100)
-                }
-            },
-            format,
+            // Delay URL revocation to ensure download completes
+            if (!url.startsWith("data:")) {
+                setTimeout(() => URL.revokeObjectURL(url), 100)
+            }
         }
 
         // Export diagram - callback will be handled in handleDiagramExport
-        drawioRef.current.exportDiagram({ format: drawioFormat })
+        drawioRef.current.exportDiagram({
+            format: drawioFormat,
+            message: tag,
+        })
     }
 
     // Log save event to Langfuse (just flags the trace, doesn't send content)
@@ -389,13 +401,14 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         <DiagramContext.Provider
             value={{
                 chartXML,
+                chartXMLRef,
                 latestSvg,
                 diagramHistory,
                 setDiagramHistory,
                 loadDiagram,
                 handleExport,
                 handleExportWithoutHistory,
-                resolverRef,
+                exportResolversRef,
                 drawioRef,
                 handleDiagramExport,
                 handleDiagramAutoSave,

docs/cn/README_CN.md

@@ -96,7 +96,6 @@ https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
 -   **图表历史记录**：全面的版本控制，跟踪所有更改，允许您查看和恢复AI编辑前的图表版本
 -   **交互式聊天界面**：与AI实时对话来完善您的图表
 -   **云架构图支持**：专门支持生成云架构图（AWS、GCP、Azure）
--   **自动计算布局**：画架构图、流程图、泳道图/BPMN、时序图、思维导图和组织架构图时，AI 只描述结构——谁包含谁，或者谁指向谁——所有坐标、尺寸和连线路径都由程序计算。容器一定装得下里面的内容，同层元素不会重叠，连线会绕开本来会穿过的图形。之后您手动移动或改色的部分会被当作图表的一部分读回来，所以后续修改不会覆盖掉您的调整。
 -   **动画连接器**：在图表元素之间创建动态动画连接器，实现更好的可视化效果
 
 ## MCP服务器
@@ -125,6 +124,13 @@ claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
 
 图表会实时显示在浏览器中！
 
+MCP服务器包含网页版的大部分画图功能：
+
+-   与网页版相同的画图规则和形状库（AWS、Azure、GCP、Kubernetes等）
+-   截图工具，AI可以查看画好的图并自行修正
+-   版本历史、多页图表，以及下载为`.drawio`、`.png`、`.svg`或`.drawio.svg`格式
+-   自动保存到`~/.next-ai-drawio/`，重启后可以接着画
+
 详情请参阅[MCP服务器README](../../packages/mcp-server/README.md)，了解VS Code、Cursor等客户端配置。
 
 ## 快速开始

docs/cn/ai-providers.md

@@ -46,6 +46,8 @@ AI_MODEL=gpt-4o
 OPENAI_BASE_URL=https://your-custom-endpoint/v1
 ```
 
+LM Studio 等本地 OpenAI 兼容服务也用同样的方式：把 base URL 设为 `http://localhost:1234/v1`（LM Studio 的默认端口），API 密钥填任意非空值即可，例如 `lm-studio`。在上面的环境变量或模型设置里配置都可以。
+
 ### AIHubMix
 
 AIHubMix 通过单个 API Key 聚合 Claude、GPT、Gemini、DeepSeek 等模型。
@@ -149,9 +151,11 @@ AZURE_REASONING_SUMMARY=detailed  # 可选：none, brief, detailed
 AWS_REGION=us-west-2
 AWS_ACCESS_KEY_ID=your_access_key_id
 AWS_SECRET_ACCESS_KEY=your_secret_access_key
-AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0
+AI_MODEL=global.anthropic.claude-sonnet-5-5
 ```
 
+较新的 Claude 模型只能通过推理配置文件（inference profile）的 ID 调用，这种 ID 以地区前缀开头，例如 `global.` 或 `us.`。直接用 `anthropic.claude-sonnet-5-5` 这样的 ID 会报错 “on-demand throughput isn't supported”。
+
 注意：在 AWS 环境（Lambda、带有 IAM 角色的 EC2）中，凭证会自动从 IAM 角色获取。
 
 ### OpenRouter
@@ -190,7 +194,7 @@ MODELSCOPE_BASE_URL=https://your-custom-endpoint
 可选的自定义 URL：
 
 ```bash
-OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_BASE_URL=http://localhost:11434/api
 ```
 
 ### Vercel AI Gateway

docs/en/ai-providers.md

@@ -61,6 +61,8 @@ Optional custom endpoint (for OpenAI-compatible services):
 OPENAI_BASE_URL=https://your-custom-endpoint/v1
 ```
 
+LM Studio and other local OpenAI-compatible servers work the same way: set the base URL to `http://localhost:1234/v1` (LM Studio's default port) and use any non-empty API key, such as `lm-studio`. This works both in the environment variables above and in the model settings.
+
 ### AIHubMix
 
 AIHubMix provides access to Claude, GPT, Gemini, DeepSeek, and other models through a single API key.
@@ -164,9 +166,11 @@ AZURE_REASONING_SUMMARY=detailed  # Optional: none, brief, detailed
 AWS_REGION=us-west-2
 AWS_ACCESS_KEY_ID=your_access_key_id
 AWS_SECRET_ACCESS_KEY=your_secret_access_key
-AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0
+AI_MODEL=global.anthropic.claude-sonnet-5-5
 ```
 
+Recent Claude models answer only through an inference profile id, which starts with a region prefix such as `global.` or `us.`. The plain id (`anthropic.claude-sonnet-5-5`) fails with "on-demand throughput isn't supported".
+
 Note: On AWS (Lambda, EC2 with IAM role), credentials are automatically obtained from the IAM role.
 
 ### OpenRouter
@@ -192,7 +196,7 @@ AI_MODEL=llama3.2
 Optional custom URL:
 
 ```bash
-OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_BASE_URL=http://localhost:11434/api
 ```
 
 ### ModelScope

docs/ja/README_JA.md

@@ -94,7 +94,6 @@ https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
 -   **ダイアグラム履歴**：すべての変更を追跡する包括的なバージョン管理。AI編集前のダイアグラムの以前のバージョンを表示・復元可能
 -   **インタラクティブなチャットインターフェース**：AIとリアルタイムでコミュニケーションしてダイアグラムを改善
 -   **クラウドアーキテクチャダイアグラムサポート**：クラウドアーキテクチャダイアグラムの生成を専門的にサポート（AWS、GCP、Azure）
--   **レイアウトの自動計算**：アーキテクチャ図、フローチャート、スイムレーン図／BPMN、シーケンス図、マインドマップ、組織図では、AI は構造だけ——何が何を含むか、何が何を指すか——を指定し、座標・サイズ・矢印の経路はすべてアプリが計算します。コンテナは必ず中身が収まるサイズになり、同じ階層の要素が重なることはなく、矢印は本来通り抜けてしまう図形を避けて引かれます。その後に手で動かしたり色を変えた部分は図の一部として読み戻されるため、次の編集で元に戻されることはありません。
 -   **アニメーションコネクタ**：より良い可視化のためにダイアグラム要素間に動的でアニメーション化されたコネクタを作成
 
 ## MCPサーバー
@@ -123,6 +122,13 @@ Claudeにダイアグラムの作成を依頼：
 
 ダイアグラムがリアルタイムでブラウザに表示されます！
 
+MCPサーバーには、Webアプリの主な作図機能が含まれています：
+
+-   Webアプリと同じ作図ルールとシェイプライブラリ（AWS、Azure、GCP、Kubernetesなど）
+-   スクリーンショットツール：AIが描画結果を確認して修正できます
+-   バージョン履歴、複数ページのダイアグラム、`.drawio`・`.png`・`.svg`・`.drawio.svg`形式でのダウンロード
+-   `~/.next-ai-drawio/`への自動保存：再起動後も続きから編集できます
+
 詳細は[MCPサーバーREADME](../../packages/mcp-server/README.md)をご覧ください（VS Code、Cursorなどのクライアント設定も含む）。
 
 ## はじめに

docs/ja/ai-providers.md

@@ -46,6 +46,8 @@ AI_MODEL=gpt-4o
 OPENAI_BASE_URL=https://your-custom-endpoint/v1
 ```
 
+LM Studio などのローカルな OpenAI 互換サーバーも同じ方法で使えます。ベース URL を `http://localhost:1234/v1`（LM Studio の既定ポート）に設定し、API キーには `lm-studio` など空でない任意の値を入力してください。上の環境変数でも、モデル設定画面でも設定できます。
+
 ### AIHubMix
 
 AIHubMix は、単一の API キーで Claude、GPT、Gemini、DeepSeek などのモデルへのアクセスを提供します。
@@ -149,9 +151,11 @@ AZURE_REASONING_SUMMARY=detailed  # 任意: none, brief, detailed
 AWS_REGION=us-west-2
 AWS_ACCESS_KEY_ID=your_access_key_id
 AWS_SECRET_ACCESS_KEY=your_secret_access_key
-AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0
+AI_MODEL=global.anthropic.claude-sonnet-5-5
 ```
 
+最近の Claude モデルは、推論プロファイル（inference profile）の ID でのみ呼び出せます。この ID は `global.` や `us.` などのリージョン接頭辞で始まります。`anthropic.claude-sonnet-5-5` のような接頭辞なしの ID では「on-demand throughput isn't supported」というエラーになります。
+
 注: AWS 上（IAM ロールを持つ Lambda や EC2）では、認証情報は IAM ロールから自動的に取得されます。
 
 ### OpenRouter
@@ -177,7 +181,7 @@ AI_MODEL=llama3.2
 任意のカスタム URL:
 
 ```bash
-OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_BASE_URL=http://localhost:11434/api
 ```
 
 ### ModelScope

docs/shape-libraries/README.md

@@ -11,7 +11,6 @@ Reference: `style="shape=mxgraph.<library>.<shape_name>"`
 | gcp2 | 297 | `mxgraph.gcp2` | Google Cloud Platform - Compute Engine, BigQuery, GKE, etc. | [gcp2.md](./gcp2.md) |
 | alibaba_cloud | 273 | `mxgraph.alibaba_cloud` | Alibaba Cloud - ECS, OSS, RDS, SLB, VPC, etc. | [alibaba_cloud.md](./alibaba_cloud.md) |
 | openstack | 18 | `mxgraph.openstack` | OpenStack cloud platform icons | [openstack.md](./openstack.md) |
-| digitalocean | 74 | `mxgraph.digitalocean` | DigitalOcean - Droplets, Spaces, Kubernetes, etc. | [digitalocean.md](./digitalocean.md) |
 | salesforce | 96 | `mxgraph.salesforce` | Salesforce platform icons | [salesforce.md](./salesforce.md) |
 
 ## Networking & Infrastructure
@@ -20,7 +19,6 @@ Reference: `style="shape=mxgraph.<library>.<shape_name>"`
 |---------|--------|--------|-------------|------|
 | cisco19 | 232 | `mxgraph.cisco19` | Cisco network equipment - routers, switches, firewalls | [cisco19.md](./cisco19.md) |
 | network | 58 | `mxgraph.networks` | General network diagram symbols | [network.md](./network.md) |
-| arista | 45 | `mxgraph.arista` | Arista network switches and equipment | [arista.md](./arista.md) |
 | kubernetes | 40 | `mxgraph.kubernetes` | Kubernetes - pods, services, deployments, nodes | [kubernetes.md](./kubernetes.md) |
 | vvd | 93 | `mxgraph.vvd` | VMware Validated Design icons | [vvd.md](./vvd.md) |
 | rack | 11 | `mxgraph.rack` | Server rack and data center equipment | [rack.md](./rack.md) |
@@ -30,7 +28,6 @@ Reference: `style="shape=mxgraph.<library>.<shape_name>"`
 | Library | Shapes | Prefix | Description | File |
 |---------|--------|--------|-------------|------|
 | bpmn | 39 | `mxgraph.bpmn` | Business Process Model and Notation - events, gateways, tasks | [bpmn.md](./bpmn.md) |
-| eip | 36 | `mxgraph.eip` | Enterprise Integration Patterns - messaging, routing | [eip.md](./eip.md) |
 | lean_mapping | 13 | `mxgraph.lean_mapping` | Lean/Value Stream Mapping symbols | [lean_mapping.md](./lean_mapping.md) |
 
 ## General Diagrams
@@ -48,6 +45,7 @@ Reference: `style="shape=mxgraph.<library>.<shape_name>"`
 | Library | Shapes | Prefix | Description | File |
 |---------|--------|--------|-------------|------|
 | android | 17 | `mxgraph.android` | Android UI mockup components | [android.md](./android.md) |
+| material_design | 300 | `image=https://fonts.gstatic.com/...` | Google Material Icons (SVG images) | [material_design.md](./material_design.md) |
 
 ## Enterprise Software
 
@@ -73,6 +71,5 @@ Reference: `style="shape=mxgraph.<library>.<shape_name>"`
 | Library | Shapes | Prefix | Description | File |
 |---------|--------|--------|-------------|------|
 | webicons | 176 | `mxgraph.webicons` | Web/social media logos - GitHub, Twitter, AWS, etc. | [webicons.md](./webicons.md) |
-| un-ocha-icons | 242 | `mxgraph.un-ocha-icons` | UN OCHA humanitarian icons | [un-ocha-icons.md](./un-ocha-icons.md) |
 
-**Total: 33 libraries, 4,281 shapes**
+**Total: 30 libraries, 4,184 shapes**

edge-functions/api/edgeai/chat/completions.ts

@@ -67,41 +67,64 @@ const MODEL_ALIASES: Record<string, string> = {
     "deepseek-v3-0324": "@tx/deepseek-ai/deepseek-v3-0324",
 }
 
-const CORS_HEADERS = {
-    "Access-Control-Allow-Origin": "*",
-    "Access-Control-Allow-Methods": "POST, OPTIONS",
-    "Access-Control-Allow-Headers": "Content-Type, Authorization",
-}
-
 /**
- * Create standardized response with CORS headers
+ * Create standardized JSON response
  */
 function createResponse(body: any, status = 200, extraHeaders = {}): Response {
     return new Response(JSON.stringify(body), {
         status,
         headers: {
             "Content-Type": "application/json",
-            ...CORS_HEADERS,
             ...extraHeaders,
         },
     })
 }
 
-/**
- * Handle OPTIONS request for CORS preflight
- */
-function handleOptionsRequest(): Response {
-    return new Response(null, {
-        headers: {
-            ...CORS_HEADERS,
-            "Access-Control-Max-Age": "86400",
-        },
-    })
+// Only the app's own server (/api/chat, /api/validate-model) calls this
+// function, so no CORS headers are sent: other sites' pages can't call it
+// from a browser and spend the deployment's Edge AI quota.
+// Same rule as lib/access-code.ts, but reading the edge function's env.
+// No codes configured (or env unavailable) means no check.
+function hasValidAccessCode(request: Request, env: any): boolean {
+    const accessCodes: string[] =
+        env?.ACCESS_CODE_LIST?.split(",")
+            .map((code: string) => code.trim())
+            .filter(Boolean) || []
+    if (accessCodes.length === 0) return true
+    const accessCode = request.headers.get("x-access-code")
+    return !!accessCode && accessCodes.includes(accessCode)
 }
 
-export async function onRequest({ request, env: _env }: any) {
-    if (request.method === "OPTIONS") {
-        return handleOptionsRequest()
+export async function onRequest({ request, env }: any) {
+    // Requiring JSON also makes any cross-site browser request need a CORS
+    // preflight, which fails without CORS headers. Only the type before any
+    // parameters counts: "text/plain; x=application/json" needs none.
+    const mediaType = (request.headers.get("content-type") ?? "")
+        .split(";")[0]
+        .trim()
+        .toLowerCase()
+    if (request.method !== "POST" || mediaType !== "application/json") {
+        return createResponse(
+            {
+                error: {
+                    message: "Expected a POST request with a JSON body",
+                    type: "invalid_request_error",
+                },
+            },
+            400,
+        )
+    }
+
+    if (!hasValidAccessCode(request, env)) {
+        return createResponse(
+            {
+                error: {
+                    message: "Invalid or missing access code",
+                    type: "invalid_request_error",
+                },
+            },
+            401,
+        )
     }
 
     request.headers.delete("accept-encoding")
@@ -153,7 +176,7 @@ export async function onRequest({ request, env: _env }: any) {
                         type: "invalid_request_error",
                     },
                 },
-                429,
+                400,
             )
         }
 
@@ -216,7 +239,6 @@ export async function onRequest({ request, env: _env }: any) {
                     "Cache-Control": "no-cache, no-store, no-transform",
                     "X-Accel-Buffering": "no",
                     Connection: "keep-alive",
-                    ...CORS_HEADERS,
                 },
             })
         } catch (error: any) {

electron/electron.d.ts

@@ -60,10 +60,6 @@ declare global {
             maximize: () => void
             /** Close the window */
             close: () => void
-            /** Open file dialog and return file path */
-            openFile: () => Promise<string | null>
-            /** Save data to file via save dialog */
-            saveFile: (data: string) => Promise<boolean>
             /** Get proxy configuration */
             getProxy: () => Promise<ProxyConfig>
             /** Set proxy configuration (saves and restarts server) */
@@ -74,6 +70,15 @@ declare global {
             >
             /** Set user's preferred locale */
             setUserLocale: (locale: string) => Promise<SetUserLocaleResult>
+            /**
+             * Call back after the server restarted on the same port (another
+             * preset); returns a function that stops the calls
+             */
+            onServerRestarted?: (callback: () => void) => () => void
+            /** A chat was saved: open this port next launch */
+            chatSaved?: () => Promise<void>
+            /** The page loaded with this many chats */
+            chatsLoaded?: (count: number) => Promise<void>
         }
 
         /** Settings window Electron API */

electron/main/app-menu.ts

@@ -32,6 +32,68 @@ export function rebuildAppMenu(): void {
     buildAppMenu()
 }
 
+// Number of the latest preset switch
+let lastSwitch = 0
+
+/**
+ * Apply a preset and restart the server so it takes effect.
+ * If the restart fails, go back to the previous preset and restart again,
+ * so the running server always matches the saved current preset.
+ * Throws an error describing the outcome on failure.
+ */
+export async function switchPreset(
+    id: string,
+): Promise<Record<string, string>> {
+    const previousPresetId = getCurrentPresetId()
+    const env = applyPresetToEnv(id)
+    if (!env) {
+        throw new Error("Preset not found")
+    }
+    const switchNumber = ++lastSwitch
+    rebuildAppMenu()
+
+    // In development, scripts/electron-dev.mjs restarts the Next.js dev server
+    if (!app.isPackaged) {
+        return env
+    }
+
+    try {
+        await restartNextServer()
+        return env
+    } catch (error) {
+        console.error("Failed to restart server:", error)
+        const reason = error instanceof Error ? error.message : String(error)
+
+        // A newer switch started meanwhile (also of this same preset), or
+        // the preset was deleted: its own restart follows, and undoing
+        // would lose that choice
+        if (switchNumber !== lastSwitch || getCurrentPresetId() !== id) {
+            throw new Error(
+                `The server could not be restarted.\n\nError: ${reason}`,
+            )
+        }
+
+        // Revert to previous preset on failure
+        if (!previousPresetId || !applyPresetToEnv(previousPresetId)) {
+            setCurrentPreset(null)
+        }
+        // Rebuild menu to restore previous checkmark state
+        rebuildAppMenu()
+
+        try {
+            await restartNextServer()
+        } catch (retryError) {
+            console.error("Failed to restart server again:", retryError)
+            throw new Error(
+                `The server could not be restarted.\n\nPlease restart the app.\n\nError: ${reason}`,
+            )
+        }
+        throw new Error(
+            `The server could not be restarted.\n\nThe previous configuration has been restored.\n\nError: ${reason}`,
+        )
+    }
+}
+
 /**
  * Get the menu template with translations
  */
@@ -192,32 +254,14 @@ function buildConfigMenu(
         type: "radio",
         checked: preset.id === currentPresetId,
         click: async () => {
-            const previousPresetId = getCurrentPresetId()
-            const env = applyPresetToEnv(preset.id)
-
-            if (env) {
-                try {
-                    await restartNextServer()
-                    rebuildAppMenu() // Rebuild menu to update checkmarks
-                } catch (error) {
-                    console.error("Failed to restart server:", error)
-
-                    // Revert to previous preset on failure
-                    if (previousPresetId) {
-                        applyPresetToEnv(previousPresetId)
-                    } else {
-                        setCurrentPreset(null)
-                    }
-
-                    // Rebuild menu to restore previous checkmark state
-                    rebuildAppMenu()
-
-                    // Show error dialog to notify user
-                    dialog.showErrorBox(
-                        "Configuration Error",
-                        `Failed to apply preset "${preset.name}". The server could not be restarted.\n\nThe previous configuration has been restored.\n\nError: ${error instanceof Error ? error.message : String(error)}`,
-                    )
-                }
+            try {
+                await switchPreset(preset.id)
+            } catch (error) {
+                // Show error dialog to notify user
+                dialog.showErrorBox(
+                    "Configuration Error",
+                    `Failed to apply preset "${preset.name}". ${error instanceof Error ? error.message : String(error)}`,
+                )
             }
         },
     }))

electron/main/config-manager.ts

@@ -1,5 +1,11 @@
 import { randomUUID } from "node:crypto"
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs"
+import {
+    existsSync,
+    mkdirSync,
+    readFileSync,
+    renameSync,
+    writeFileSync,
+} from "node:fs"
 import path from "node:path"
 import { app, safeStorage } from "electron"
 
@@ -30,7 +36,9 @@ let hasWarnedAboutPlaintext = false
  * Warns if encryption is not available (API key stored in plaintext)
  */
 function encryptValue(value: string): string {
-    if (!value) {
+    // Already encrypted (a value that could not be decrypted): keep it as is
+    // instead of wrapping it in a second layer of encryption
+    if (!value || value.startsWith(ENCRYPTED_PREFIX)) {
         return value
     }
 
@@ -61,6 +69,7 @@ function encryptValue(value: string): string {
 /**
  * Decrypt a sensitive value using safeStorage
  * Returns the original value if it's not encrypted or decryption fails
+ * (so saving writes the stored ciphertext back unchanged)
  */
 function decryptValue(value: string): string {
     if (!value || !value.startsWith(ENCRYPTED_PREFIX)) {
@@ -150,6 +159,10 @@ function getConfigFilePath(): string {
     return path.join(userDataPath, CONFIG_FILE_NAME)
 }
 
+// The presets file exists but the last read failed: a save now would
+// replace the user's presets with the empty list that read returned
+let presetsUnreadable = false
+
 /**
  * Load presets from the config file
  * Decrypts sensitive fields automatically
@@ -158,6 +171,25 @@ export function loadPresets(): ConfigPresetsFile {
     const configPath = getConfigFilePath()
 
     if (!existsSync(configPath)) {
+        // Nothing left that a save could overwrite
+        presetsUnreadable = false
+        return {
+            version: 1,
+            currentPresetId: null,
+            presets: [],
+            userLocale: undefined,
+        }
+    }
+
+    let content: string
+    try {
+        content = readFileSync(configPath, "utf-8")
+        presetsUnreadable = false
+    } catch (error) {
+        // Often only for now (on Windows an antivirus scanner can hold the
+        // file): keep the file, and refuse saves based on this empty list
+        console.error("Failed to read config presets:", error)
+        presetsUnreadable = true
         return {
             version: 1,
             currentPresetId: null,
@@ -167,7 +199,6 @@ export function loadPresets(): ConfigPresetsFile {
     }
 
     try {
-        const content = readFileSync(configPath, "utf-8")
         const data = JSON.parse(content) as ConfigPresetsFile
 
         // Decrypt sensitive fields in each preset
@@ -179,6 +210,17 @@ export function loadPresets(): ConfigPresetsFile {
         return data
     } catch (error) {
         console.error("Failed to load config presets:", error)
+        // Move the unreadable file aside so the next save can't overwrite
+        // the user's presets with an empty list
+        const backupPath = `${configPath}.corrupt-${Date.now()}`
+        try {
+            renameSync(configPath, backupPath)
+            console.error(`Unreadable config presets moved to ${backupPath}`)
+        } catch (renameError) {
+            // Still there: refuse saves that would overwrite it
+            console.error("Failed to back up config presets:", renameError)
+            presetsUnreadable = true
+        }
         return {
             version: 1,
             currentPresetId: null,
@@ -193,6 +235,11 @@ export function loadPresets(): ConfigPresetsFile {
  * Encrypts sensitive fields automatically
  */
 export function savePresets(data: ConfigPresetsFile): void {
+    if (presetsUnreadable) {
+        throw new Error(
+            "The presets file could not be read, so it was not overwritten. Please try again.",
+        )
+    }
     const configPath = getConfigFilePath()
     const userDataPath = app.getPath("userData")
 
@@ -211,7 +258,11 @@ export function savePresets(data: ConfigPresetsFile): void {
     }
 
     try {
-        writeFileSync(configPath, JSON.stringify(dataToSave, null, 2), "utf-8")
+        // Write a temp file and rename it, so a crash mid-write can't leave
+        // a truncated config file
+        const tempPath = `${configPath}.tmp`
+        writeFileSync(tempPath, JSON.stringify(dataToSave, null, 2), "utf-8")
+        renameSync(tempPath, configPath)
     } catch (error) {
         console.error("Failed to save config presets:", error)
         throw error
@@ -307,9 +358,10 @@ export function deletePreset(id: string): boolean {
 
     data.presets.splice(index, 1)
 
-    // Clear current preset if it was deleted
+    // Clear current preset (and its env vars) if it was deleted
     if (data.currentPresetId === id) {
         data.currentPresetId = null
+        setPresetEnv(null)
     }
 
     savePresets(data)
@@ -322,13 +374,15 @@ export function deletePreset(id: string): boolean {
 export function setCurrentPreset(id: string | null): boolean {
     const data = loadPresets()
 
+    let preset: ConfigPreset | null = null
     if (id !== null) {
-        const preset = data.presets.find((p) => p.id === id)
+        preset = data.presets.find((p) => p.id === id) || null
         if (!preset) {
             return false
         }
     }
 
+    setPresetEnv(preset)
     data.currentPresetId = id
     savePresets(data)
     return true
@@ -365,78 +419,23 @@ const PROVIDER_ENV_MAP: Record<string, { apiKey: string; baseUrl: string }> = {
 }
 
 /**
- * Apply preset environment variables to the current process
- * Returns the environment variables that were applied
- */
-export function applyPresetToEnv(id: string): Record<string, string> | null {
-    const data = loadPresets()
-    const preset = data.presets.find((p) => p.id === id)
-
-    if (!preset) {
-        return null
-    }
-
-    const appliedEnv: Record<string, string> = {}
-    const provider = preset.config.AI_PROVIDER?.toLowerCase()
-
-    for (const [key, value] of Object.entries(preset.config)) {
-        if (value !== undefined && value !== "") {
-            // Map generic AI_API_KEY to provider-specific key
-            if (
-                key === "AI_API_KEY" &&
-                provider &&
-                PROVIDER_ENV_MAP[provider]
-            ) {
-                const providerApiKey = PROVIDER_ENV_MAP[provider].apiKey
-                if (providerApiKey) {
-                    process.env[providerApiKey] = value
-                    appliedEnv[providerApiKey] = value
-                }
-            }
-            // Map generic AI_BASE_URL to provider-specific key
-            else if (
-                key === "AI_BASE_URL" &&
-                provider &&
-                PROVIDER_ENV_MAP[provider]
-            ) {
-                const providerBaseUrl = PROVIDER_ENV_MAP[provider].baseUrl
-                if (providerBaseUrl) {
-                    process.env[providerBaseUrl] = value
-                    appliedEnv[providerBaseUrl] = value
-                }
-            }
-            // Apply other env vars directly
-            else {
-                process.env[key] = value
-                appliedEnv[key] = value
-            }
-        }
-    }
-
-    // Set as current preset
-    data.currentPresetId = id
-    savePresets(data)
-
-    return appliedEnv
-}
-
-/**
- * Get environment variables from current preset
+ * Map a preset's config to environment variables
  * Maps generic AI_API_KEY/AI_BASE_URL to provider-specific keys
  */
-export function getCurrentPresetEnv(): Record<string, string> {
-    const preset = getCurrentPreset()
-    if (!preset) {
-        return {}
-    }
-
+function presetToEnv(preset: ConfigPreset): Record<string, string> {
     const env: Record<string, string> = {}
     const provider = preset.config.AI_PROVIDER?.toLowerCase()
 
     for (const [key, value] of Object.entries(preset.config)) {
         if (value !== undefined && value !== "") {
+            // A key that could not be decrypted is useless to the server
+            if (value.startsWith(ENCRYPTED_PREFIX)) {
+                console.warn(
+                    `Preset "${preset.name}": ${key} could not be decrypted. Please enter it again in Settings.`,
+                )
+            }
             // Map generic AI_API_KEY to provider-specific key
-            if (
+            else if (
                 key === "AI_API_KEY" &&
                 provider &&
                 PROVIDER_ENV_MAP[provider]
@@ -466,6 +465,90 @@ export function getCurrentPresetEnv(): Record<string, string> {
     return env
 }
 
+/**
+ * Values that env vars had before a preset first set them
+ * (from the system or .env files), and the keys the active preset set
+ */
+const originalEnv: Record<string, string | undefined> = {}
+let presetEnvKeys: string[] = []
+
+/**
+ * Replace the env vars of the previous preset with those of the given preset
+ * (null leaves no preset applied). Restoring first means switching presets
+ * never leaves the previous preset's base URL, model or key behind.
+ */
+function setPresetEnv(preset: ConfigPreset | null): Record<string, string> {
+    for (const key of presetEnvKeys) {
+        if (originalEnv[key] === undefined) {
+            delete process.env[key]
+        } else {
+            process.env[key] = originalEnv[key]
+        }
+    }
+
+    const env = preset ? presetToEnv(preset) : {}
+    for (const [key, value] of Object.entries(env)) {
+        if (!(key in originalEnv)) {
+            originalEnv[key] = process.env[key]
+        }
+        process.env[key] = value
+    }
+    presetEnvKeys = Object.keys(env)
+
+    writeDevPresetEnv(env)
+    return env
+}
+
+const DEV_ENV_FILE_NAME = "dev-preset-env.json"
+
+/**
+ * Development only: write the active preset's env vars (decrypted and mapped)
+ * for scripts/electron-dev.mjs, which restarts the Next.js dev server when
+ * this file changes. The dev server can't decrypt the config file itself.
+ */
+function writeDevPresetEnv(env: Record<string, string>): void {
+    if (app.isPackaged) {
+        return
+    }
+    try {
+        const filePath = path.join(app.getPath("userData"), DEV_ENV_FILE_NAME)
+        writeFileSync(filePath, JSON.stringify(env, null, 2), {
+            encoding: "utf-8",
+            mode: 0o600,
+        })
+    } catch (error) {
+        console.error("Failed to write dev preset env:", error)
+    }
+}
+
+/**
+ * Apply preset environment variables to the current process
+ * Returns the environment variables that were applied
+ */
+export function applyPresetToEnv(id: string): Record<string, string> | null {
+    const data = loadPresets()
+    const preset = data.presets.find((p) => p.id === id)
+
+    if (!preset) {
+        return null
+    }
+
+    const appliedEnv = setPresetEnv(preset)
+
+    // Set as current preset
+    data.currentPresetId = id
+    savePresets(data)
+
+    return appliedEnv
+}
+
+/**
+ * Apply the saved current preset's environment variables (used at startup)
+ */
+export function applyCurrentPresetToEnv(): void {
+    setPresetEnv(getCurrentPreset())
+}
+
 /**
  * Get user's preferred locale from config
  * Returns undefined if not set

electron/main/env-loader.ts

@@ -28,6 +28,25 @@ export function loadEnvFile(): void {
     console.log("No .env file found, using system environment variables")
 }
 
+/**
+ * Index of the quote that closes a value starting with a quote, or -1. A
+ * backslash before the quote character escapes it, as in dotenv; the
+ * backslash stays in the value. As in dotenv, an escaped quote with only a
+ * comment or nothing after it still closes the value when no other quote
+ * does ("C:\dir\" keeps its trailing backslash).
+ */
+function findClosingQuote(value: string): number {
+    const quote = value[0]
+    let lastEscaped = -1
+    for (let i = 1; i < value.length; i++) {
+        if (value[i] === "\\" && value[i + 1] === quote) {
+            i++
+            if (/^\s*(#.*)?$/.test(value.slice(i + 1))) lastEscaped = i
+        } else if (value[i] === quote) return i
+    }
+    return lastEscaped
+}
+
 /**
  * Parse and load environment variables from a file
  */
@@ -48,12 +67,28 @@ function loadEnvFromFile(filePath: string): void {
             const key = trimmed.slice(0, equalIndex).trim()
             let value = trimmed.slice(equalIndex + 1).trim()
 
-            // Remove surrounding quotes
+            const quote = value[0]
+            const closingQuote =
+                quote === '"' || quote === "'" ? findClosingQuote(value) : -1
             if (
-                (value.startsWith('"') && value.endsWith('"')) ||
-                (value.startsWith("'") && value.endsWith("'"))
+                closingQuote > 0 &&
+                /^\s*(#.*)?$/.test(value.slice(closingQuote + 1))
             ) {
-                value = value.slice(1, -1)
+                // Quoted value, then nothing or a comment: keep what is
+                // inside the quotes, as dotenv reads it
+                value = value.slice(1, closingQuote)
+            } else {
+                // Unquoted value: drop an inline comment ("value  # comment").
+                // A value quoted from start to end with quotes inside (JSON
+                // with an apostrophe) loses only the outer two, as in dotenv.
+                value = value.replace(/\s+#.*$/, "")
+                if (
+                    closingQuote > 0 &&
+                    value.length > 1 &&
+                    value.endsWith(quote)
+                ) {
+                    value = value.slice(1, -1)
+                }
             }
 
             // Don't override existing environment variables

electron/main/index.ts

@@ -1,12 +1,17 @@
 import { app, BrowserWindow, dialog, shell } from "electron"
 import { buildAppMenu } from "./app-menu"
-import { getCurrentPresetEnv } from "./config-manager"
+import { applyCurrentPresetToEnv } from "./config-manager"
 import { loadEnvFile } from "./env-loader"
 import { registerIpcHandlers } from "./ipc-handlers"
 import { startNextServer, stopNextServer } from "./next-server"
 import { applyProxyToEnv } from "./proxy-manager"
 import { registerSettingsWindowHandlers } from "./settings-window"
-import { createWindow, getMainWindow } from "./window-manager"
+import {
+    createWindow,
+    getAppUrl,
+    getMainWindow,
+    isAppUrl,
+} from "./window-manager"
 
 // Single instance lock
 const gotTheLock = app.requestSingleInstanceLock()
@@ -28,16 +33,14 @@ if (!gotTheLock) {
     // Apply proxy settings from saved config
     applyProxyToEnv()
 
-    // Apply saved preset environment variables (overrides .env)
-    const presetEnv = getCurrentPresetEnv()
-    for (const [key, value] of Object.entries(presetEnv)) {
-        process.env[key] = value
-    }
-
-    const isDev = process.env.NODE_ENV === "development"
-    let serverUrl: string | null = null
+    const isDev = !app.isPackaged
 
     app.whenReady().then(async () => {
+        // Apply saved preset environment variables (overrides .env).
+        // Must run after ready: on Windows and Linux safeStorage can't
+        // decrypt the API key before that.
+        applyCurrentPresetToEnv()
+
         // Register IPC handlers
         registerIpcHandlers()
         registerSettingsWindowHandlers()
@@ -46,6 +49,7 @@ if (!gotTheLock) {
         buildAppMenu()
 
         try {
+            let serverUrl: string
             if (isDev) {
                 // Development: use the dev server URL
                 serverUrl =
@@ -69,8 +73,9 @@ if (!gotTheLock) {
 
         app.on("activate", () => {
             if (BrowserWindow.getAllWindows().length === 0) {
-                if (serverUrl) {
-                    createWindow(serverUrl)
+                const appUrl = getAppUrl()
+                if (appUrl) {
+                    createWindow(appUrl)
                 }
             }
         })
@@ -87,24 +92,49 @@ if (!gotTheLock) {
         stopNextServer()
     })
 
+    // Pages allowed inside app windows: the app server and draw.io
+    const isInAppUrl = (url: string): boolean => {
+        if (isAppUrl(url)) return true
+        try {
+            const { hostname } = new URL(url)
+            return ["diagrams.net", "draw.io"].some(
+                (domain) =>
+                    hostname === domain || hostname.endsWith(`.${domain}`),
+            )
+        } catch {
+            return false
+        }
+    }
+
+    const isWebUrl = (url: string): boolean =>
+        url.startsWith("http://") || url.startsWith("https://")
+
     // Open external links in default browser
     app.on("web-contents-created", (_, contents) => {
         contents.setWindowOpenHandler(({ url }) => {
-            // Allow diagrams.net iframe
-            if (
-                url.includes("diagrams.net") ||
-                url.includes("draw.io") ||
-                url.startsWith("http://localhost") ||
-                url.startsWith("http://127.0.0.1")
-            ) {
+            if (isInAppUrl(url)) {
                 return { action: "allow" }
             }
             // Open other links in external browser
-            if (url.startsWith("http://") || url.startsWith("https://")) {
+            if (isWebUrl(url)) {
                 shell.openExternal(url)
                 return { action: "deny" }
             }
             return { action: "allow" }
         })
+
+        // Clicking a plain link would otherwise replace the app page with
+        // an external site that keeps the preload API. Only the page
+        // itself may navigate there; draw.io stays in its frame (this event
+        // is for the main frame only)
+        contents.on("will-navigate", (event) => {
+            if (isAppUrl(event.url)) {
+                return
+            }
+            event.preventDefault()
+            if (isWebUrl(event.url)) {
+                shell.openExternal(event.url)
+            }
+        })
     })
 }

electron/main/ipc-handlers.ts

@@ -1,7 +1,6 @@
-import { app, BrowserWindow, dialog, ipcMain } from "electron"
-import { rebuildAppMenu } from "./app-menu"
+import { app, BrowserWindow, type IpcMainInvokeEvent, ipcMain } from "electron"
+import { rebuildAppMenu, switchPreset } from "./app-menu"
 import {
-    applyPresetToEnv,
     type ConfigPreset,
     createPreset,
     deletePreset,
@@ -14,12 +13,14 @@ import {
     updatePreset,
 } from "./config-manager"
 import { restartNextServer } from "./next-server"
+import { noteNoChats, rememberChatPort } from "./port-manager"
 import {
     applyProxyToEnv,
     getProxyConfig,
     type ProxyConfig,
     saveProxyConfig,
 } from "./proxy-manager"
+import { isAppUrl } from "./window-manager"
 
 /**
  * Allowed configuration keys for presets
@@ -48,16 +49,44 @@ function sanitizePresetConfig(
     return sanitized
 }
 
+/**
+ * Register an IPC handler that only answers the app's own pages
+ * (the main window on the app server, or the local settings page).
+ * A main window that somehow ends up on an external site still gets the
+ * preload API, so its calls must be rejected here.
+ */
+function handle<Args extends unknown[]>(
+    channel: string,
+    listener: (event: IpcMainInvokeEvent, ...args: Args) => unknown,
+): void {
+    ipcMain.handle(channel, (event, ...args) => {
+        const url = event.senderFrame?.url
+        if (!isAppUrl(url) && !url?.startsWith("file://")) {
+            throw new Error(`Blocked "${channel}" from untrusted page: ${url}`)
+        }
+        return listener(event, ...(args as Args))
+    })
+}
+
 /**
  * Register all IPC handlers
  */
 export function registerIpcHandlers(): void {
     // ==================== App Info ====================
 
-    ipcMain.handle("get-version", () => {
+    handle("get-version", () => {
         return app.getVersion()
     })
 
+    // ==================== Where the chats are ====================
+
+    // The page saved a chat, or loaded without any: decides which port
+    // (and so which origin's chats) the next launch opens
+    handle("chat-saved", () => rememberChatPort())
+    handle("chats-loaded", (_event, count: unknown) => {
+        if (count === 0) noteNoChats()
+    })
+
     // ==================== Window Controls ====================
 
     ipcMain.on("window-minimize", (event) => {
@@ -79,84 +108,30 @@ export function registerIpcHandlers(): void {
         win?.close()
     })
 
-    // ==================== File Dialogs ====================
-
-    ipcMain.handle("dialog-open-file", async (event) => {
-        const win = BrowserWindow.fromWebContents(event.sender)
-        if (!win) return null
-
-        const result = await dialog.showOpenDialog(win, {
-            properties: ["openFile"],
-            filters: [
-                { name: "Draw.io Files", extensions: ["drawio", "xml"] },
-                { name: "All Files", extensions: ["*"] },
-            ],
-        })
-
-        if (result.canceled || result.filePaths.length === 0) {
-            return null
-        }
-
-        // Read the file content
-        const fs = await import("node:fs/promises")
-        try {
-            const content = await fs.readFile(result.filePaths[0], "utf-8")
-            return content
-        } catch (error) {
-            console.error("Failed to read file:", error)
-            return null
-        }
-    })
-
-    ipcMain.handle("dialog-save-file", async (event, data: string) => {
-        const win = BrowserWindow.fromWebContents(event.sender)
-        if (!win) return false
-
-        const result = await dialog.showSaveDialog(win, {
-            filters: [
-                { name: "Draw.io Files", extensions: ["drawio"] },
-                { name: "XML Files", extensions: ["xml"] },
-            ],
-        })
-
-        if (result.canceled || !result.filePath) {
-            return false
-        }
-
-        const fs = await import("node:fs/promises")
-        try {
-            await fs.writeFile(result.filePath, data, "utf-8")
-            return true
-        } catch (error) {
-            console.error("Failed to save file:", error)
-            return false
-        }
-    })
-
     // ==================== Config Presets ====================
 
-    ipcMain.handle("config-presets:get-all", () => {
+    handle("config-presets:get-all", () => {
         return getAllPresets()
     })
 
-    ipcMain.handle("config-presets:get-current", () => {
+    handle("config-presets:get-current", () => {
         return getCurrentPreset()
     })
 
-    ipcMain.handle("config-presets:get-current-id", () => {
+    handle("config-presets:get-current-id", () => {
         return getCurrentPresetId()
     })
 
-    ipcMain.handle(
+    handle(
         "config-presets:save",
-        (
+        async (
             _event,
             preset: Omit<ConfigPreset, "id" | "createdAt" | "updatedAt"> & {
                 id?: string
             },
         ) => {
             // Validate preset name
-            if (typeof preset.name !== "string" || !preset.name.trim()) {
+            if (typeof preset?.name !== "string" || !preset.name.trim()) {
                 throw new Error("Invalid preset name")
             }
 
@@ -165,42 +140,48 @@ export function registerIpcHandlers(): void {
 
             if (preset.id) {
                 // Update existing preset
-                return updatePreset(preset.id, {
+                const updated = updatePreset(preset.id, {
                     name: preset.name.trim(),
                     config: sanitizedConfig,
                 })
+                // Re-apply the active preset so the edit takes effect
+                if (updated && updated.id === getCurrentPresetId()) {
+                    await switchPreset(updated.id)
+                } else {
+                    rebuildAppMenu()
+                }
+                return updated
             }
             // Create new preset
-            return createPreset({
+            const created = createPreset({
                 name: preset.name.trim(),
                 config: sanitizedConfig,
             })
+            rebuildAppMenu()
+            return created
         },
     )
 
-    ipcMain.handle("config-presets:delete", (_event, id: string) => {
-        return deletePreset(id)
+    handle("config-presets:delete", async (_event, id: string) => {
+        const wasCurrent = id === getCurrentPresetId()
+        // Deleting the active preset also clears its env vars
+        const deleted = deletePreset(id)
+        rebuildAppMenu()
+
+        // Restart so the server stops using the deleted preset
+        if (deleted && wasCurrent && app.isPackaged) {
+            await restartNextServer()
+        }
+        return deleted
     })
 
-    ipcMain.handle("config-presets:apply", async (_event, id: string) => {
-        const env = applyPresetToEnv(id)
-        if (!env) {
-            return { success: false, error: "Preset not found" }
-        }
-
-        const isDev = process.env.NODE_ENV === "development"
-
-        if (isDev) {
-            // In development mode, the config file change will trigger
-            // the file watcher in electron-dev.mjs to restart Next.js
-            // We just need to save the preset (already done in applyPresetToEnv)
-            return { success: true, env, devMode: true }
-        }
-
-        // Production mode: restart the Next.js server to apply new environment variables
+    handle("config-presets:apply", async (_event, id: string) => {
         try {
-            await restartNextServer()
-            return { success: true, env }
+            const env = await switchPreset(id)
+            // In development mode, electron-dev.mjs restarts Next.js
+            return app.isPackaged
+                ? { success: true, env }
+                : { success: true, env, devMode: true }
         } catch (error) {
             return {
                 success: false,
@@ -212,30 +193,39 @@ export function registerIpcHandlers(): void {
         }
     })
 
-    ipcMain.handle(
-        "config-presets:set-current",
-        (_event, id: string | null) => {
-            return setCurrentPreset(id)
-        },
-    )
+    handle("config-presets:set-current", (_event, id: string | null) => {
+        return setCurrentPreset(id)
+    })
 
     // ==================== Proxy Settings ====================
 
-    ipcMain.handle("get-proxy", () => {
+    handle("get-proxy", () => {
         return getProxyConfig()
     })
 
-    ipcMain.handle("set-proxy", async (_event, config: ProxyConfig) => {
+    handle("set-proxy", async (_event, config: ProxyConfig) => {
+        const isOptionalString = (value: unknown) =>
+            value === undefined || typeof value === "string"
+        if (
+            typeof config !== "object" ||
+            config === null ||
+            !isOptionalString(config.httpProxy) ||
+            !isOptionalString(config.httpsProxy)
+        ) {
+            return { success: false, error: "Invalid proxy settings" }
+        }
+
         try {
             // Save config to file
-            saveProxyConfig(config)
+            saveProxyConfig({
+                httpProxy: config.httpProxy,
+                httpsProxy: config.httpsProxy,
+            })
 
             // Apply to current process environment
             applyProxyToEnv()
 
-            const isDev = process.env.NODE_ENV === "development"
-
-            if (isDev) {
+            if (!app.isPackaged) {
                 // In development, env vars are already applied
                 // Next.js dev server may need manual restart
                 return { success: true, devMode: true }
@@ -257,11 +247,11 @@ export function registerIpcHandlers(): void {
 
     // ==================== User Locale ====================
 
-    ipcMain.handle("get-user-locale", () => {
+    handle("get-user-locale", () => {
         return getUserLocale()
     })
 
-    ipcMain.handle("set-user-locale", (_event, locale: string) => {
+    handle("set-user-locale", (_event, locale: string) => {
         // Validate locale is one of the supported values
         if (!["en", "zh", "ja", "zh-Hant"].includes(locale)) {
             return { success: false, error: "Invalid locale" }

electron/main/next-server.ts

@@ -7,9 +7,20 @@ import {
     getServerUrl,
     isPortAvailable,
 } from "./port-manager"
+import { setAppUrl } from "./window-manager"
 
 let serverProcess: UtilityProcess | null = null
 
+// Start and restart run one at a time, so overlapping calls (e.g. two quick
+// preset switches) can't leave two servers running
+let serverQueue: Promise<unknown> = Promise.resolve()
+
+function runExclusive<T>(task: () => Promise<T>): Promise<T> {
+    const result = serverQueue.then(task)
+    serverQueue = result.catch(() => {})
+    return result
+}
+
 /**
  * Get the path to the standalone server resources
  * In packaged app: resources/standalone
@@ -45,7 +56,11 @@ async function waitForServer(url: string, timeout = 30000): Promise<void> {
  * Start the Next.js standalone server using Electron's utilityProcess
  * This API is designed for running Node.js code in the background
  */
-export async function startNextServer(): Promise<string> {
+export function startNextServer(): Promise<string> {
+    return runExclusive(startServer)
+}
+
+async function startServer(): Promise<string> {
     const resourcePath = getResourcePath()
     const serverPath = path.join(resourcePath, "server.js")
 
@@ -71,6 +86,13 @@ export async function startNextServer(): Promise<string> {
         HOSTNAME: "127.0.0.1",
         // Enable Node.js built-in proxy support for fetch (Node.js 24+)
         NODE_USE_ENV_PROXY: "1",
+        // The preset keys are the user's own, not a server's
+        NEXT_AI_DRAWIO_DESKTOP: "1",
+    }
+
+    // Keep requests to local model servers (e.g. Ollama) off the proxy
+    if (!process.env.NO_PROXY && !process.env.no_proxy) {
+        env.NO_PROXY = "localhost,127.0.0.1,[::1]"
     }
 
     // Set cache directory to a writable location (user's app data folder)
@@ -96,23 +118,27 @@ export async function startNextServer(): Promise<string> {
 
     // Use Electron's utilityProcess API for running Node.js in background
     // This is the recommended way to run Node.js code in Electron
-    serverProcess = utilityProcess.fork(serverPath, [], {
+    const proc = utilityProcess.fork(serverPath, [], {
         cwd: resourcePath,
         env,
         stdio: "pipe",
     })
+    serverProcess = proc
 
-    serverProcess.stdout?.on("data", (data) => {
+    proc.stdout?.on("data", (data) => {
         console.log(`[Next.js] ${data.toString().trim()}`)
     })
 
-    serverProcess.stderr?.on("data", (data) => {
+    proc.stderr?.on("data", (data) => {
         console.error(`[Next.js Error] ${data.toString().trim()}`)
     })
 
-    serverProcess.on("exit", (code) => {
+    proc.on("exit", (code) => {
         console.log(`Next.js server exited with code ${code}`)
-        serverProcess = null
+        // An old server can exit after a new one started; keep the new one
+        if (serverProcess === proc) {
+            serverProcess = null
+        }
     })
 
     const url = getServerUrl()
@@ -126,39 +152,36 @@ export async function startNextServer(): Promise<string> {
  * Stop the Next.js server process and wait for it to exit
  */
 export async function stopNextServer(): Promise<void> {
-    if (serverProcess) {
-        console.log("Stopping Next.js server...")
+    const proc = serverProcess
+    if (!proc) {
+        return
+    }
+    console.log("Stopping Next.js server...")
+    serverProcess = null
 
-        // Create a promise that resolves when the process exits
-        const exitPromise = new Promise<void>((resolve) => {
-            const proc = serverProcess
-            if (!proc) {
-                resolve()
-                return
-            }
-
-            const onExit = () => {
-                resolve()
-            }
-
-            proc.once("exit", onExit)
-
-            // Timeout after 5 seconds
-            setTimeout(() => {
-                proc.removeListener("exit", onExit)
-                resolve()
-            }, 5000)
+    // Resolves true when the process exits, false after the timeout
+    const waitForExit = (ms: number) =>
+        new Promise<boolean>((resolve) => {
+            proc.once("exit", () => resolve(true))
+            setTimeout(() => resolve(false), ms)
         })
 
-        serverProcess.kill()
-        serverProcess = null
+    proc.kill()
 
-        // Wait for process to exit
-        await exitPromise
-
-        // Additional wait for OS to release port
-        await new Promise((resolve) => setTimeout(resolve, 500))
+    // Next.js waits for open requests (e.g. a streaming reply) before it
+    // exits, so force kill it if it is still running after 5 seconds
+    if (!(await waitForExit(5000)) && proc.pid) {
+        console.warn("Next.js server did not exit in time, force killing it")
+        try {
+            process.kill(proc.pid, "SIGKILL")
+        } catch (error) {
+            console.error("Failed to force kill Next.js server:", error)
+        }
+        await waitForExit(2000)
     }
+
+    // Additional wait for OS to release port
+    await new Promise((resolve) => setTimeout(resolve, 500))
 }
 
 /**
@@ -184,15 +207,19 @@ async function waitForServerStop(timeout = 5000): Promise<void> {
 /**
  * Restart the Next.js server with new environment variables
  */
-export async function restartNextServer(): Promise<string> {
-    console.log("Restarting Next.js server...")
+export function restartNextServer(): Promise<string> {
+    return runExclusive(async () => {
+        console.log("Restarting Next.js server...")
 
-    // Stop the current server and wait for it to exit
-    await stopNextServer()
+        // Stop the current server and wait for it to exit
+        await stopNextServer()
 
-    // Wait for the port to be released
-    await waitForServerStop()
+        // Wait for the port to be released
+        await waitForServerStop()
 
-    // Start the server again
-    return startNextServer()
+        // Start the server again, and follow it if it moved to another port
+        const url = await startServer()
+        setAppUrl(url)
+        return url
+    })
 }

electron/main/port-manager.ts

@@ -1,4 +1,6 @@
+import { existsSync, readFileSync, writeFileSync } from "node:fs"
 import net from "node:net"
+import path from "node:path"
 import { app } from "electron"
 
 /**
@@ -23,6 +25,69 @@ const PORT_CONFIG = {
  */
 let allocatedPort: number | null = null
 
+/**
+ * Whether chats are saved under http://127.0.0.1:<port>: Electron keeps
+ * each origin's IndexedDB in its own folder
+ */
+function hasStoredData(port: number): boolean {
+    return existsSync(
+        path.join(
+            app.getPath("userData"),
+            "IndexedDB",
+            `http_127.0.0.1_${port}.indexeddb.leveldb`,
+        ),
+    )
+}
+
+// The two fixed production ports, the only ones whose origin (and so its
+// chats and settings) is the same at every launch
+const HOME_PORTS = [PORT_CONFIG.legacyProduction, PORT_CONFIG.production]
+
+const chatPortFile = () => path.join(app.getPath("userData"), "chat-port.json")
+
+/** The fixed port where a chat was last saved, if known */
+function readChatPort(): number | null {
+    try {
+        const { port } = JSON.parse(readFileSync(chatPortFile(), "utf-8"))
+        return HOME_PORTS.includes(port) ? port : null
+    } catch {
+        return null
+    }
+}
+
+function writeChatPort(port: number): void {
+    try {
+        writeFileSync(chatPortFile(), JSON.stringify({ port }))
+    } catch (error) {
+        console.warn("Could not save the chat port:", error)
+    }
+}
+
+/**
+ * The page saved a chat: open on this port next time. Chats of the two
+ * ports cannot be shown together (each origin has its own storage), so the
+ * app opens where the user last worked. A launch that had to use the other
+ * port and saved nothing does not move it.
+ */
+export function rememberChatPort(): void {
+    const port = allocatedPort
+    if (!app.isPackaged || port === null || !HOME_PORTS.includes(port)) return
+    if (readChatPort() !== port) writeChatPort(port)
+}
+
+/**
+ * The page loaded without any chats. Before any chat was saved under this
+ * version (no file yet), the user's chats may be on the other fixed port,
+ * where an older version opened: try it first next time.
+ */
+export function noteNoChats(): void {
+    const port = allocatedPort
+    if (!app.isPackaged || port === null || !HOME_PORTS.includes(port)) return
+    if (existsSync(chatPortFile())) return
+    const other = HOME_PORTS.find((p) => p !== port)
+    if (other !== undefined && hasStoredData(other)) writeChatPort(other)
+}
+
 /**
  * Check if a specific port is available
  */
@@ -44,7 +109,8 @@ export function isPortAvailable(port: number): Promise<boolean> {
 /**
  * Find an available port
  * - In development: uses fixed port (6002)
- * - In production: uses fixed port (13370) to preserve localStorage
+ * - In production: uses the legacy port (61337), then 13370, to preserve
+ *   localStorage; 13370 first when only it has saved chats
  * - Falls back to sequential ports if preferred port is unavailable
  * - Last resort: lets the OS assign a port (port 0)
  *
@@ -69,21 +135,26 @@ export async function findAvailablePort(reuseExisting = true): Promise<number> {
         allocatedPort = null
     }
 
-    // In production, try legacy port first to preserve existing users' localStorage
-    if (!isDev) {
-        const legacyPort = PORT_CONFIG.legacyProduction
-        if (await isPortAvailable(legacyPort)) {
-            allocatedPort = legacyPort
-            return legacyPort
+    // In production, first the port where a chat was last saved. Without
+    // one, the legacy port first to preserve existing users' data, unless
+    // only the new port has data: their app started on 13370 while Windows
+    // reserved 61337, and 61337 being free now would hide it
+    const chatPort = isDev ? null : readChatPort()
+    const candidates = isDev
+        ? [preferredPort]
+        : chatPort !== null
+          ? [chatPort, ...HOME_PORTS.filter((p) => p !== chatPort)]
+          : hasStoredData(PORT_CONFIG.production) &&
+              !hasStoredData(PORT_CONFIG.legacyProduction)
+            ? [PORT_CONFIG.production, PORT_CONFIG.legacyProduction]
+            : [PORT_CONFIG.legacyProduction, PORT_CONFIG.production]
+    for (const port of candidates) {
+        if (await isPortAvailable(port)) {
+            allocatedPort = port
+            return port
         }
     }
 
-    // Try preferred port
-    if (await isPortAvailable(preferredPort)) {
-        allocatedPort = preferredPort
-        return preferredPort
-    }
-
     console.warn(
         `Preferred port ${preferredPort} is in use, finding alternative...`,
     )

electron/main/proxy-manager.ts

@@ -13,18 +13,22 @@ function getConfigPath(): string {
 
 /**
  * Load proxy configuration from JSON file
+ * Returns null if the user never saved proxy settings (or the file is invalid)
  */
-export function loadProxyConfig(): ProxyConfig {
+export function loadProxyConfig(): ProxyConfig | null {
     try {
         const configPath = getConfigPath()
         if (fs.existsSync(configPath)) {
-            const data = fs.readFileSync(configPath, "utf-8")
-            return JSON.parse(data) as ProxyConfig
+            const data = JSON.parse(fs.readFileSync(configPath, "utf-8"))
+            if (data && typeof data === "object" && !Array.isArray(data)) {
+                return data as ProxyConfig
+            }
+            console.error("Ignoring invalid proxy config:", data)
         }
     } catch (error) {
         console.error("Failed to load proxy config:", error)
     }
-    return {}
+    return null
 }
 
 /**
@@ -33,7 +37,11 @@ export function loadProxyConfig(): ProxyConfig {
 export function saveProxyConfig(config: ProxyConfig): void {
     try {
         const configPath = getConfigPath()
-        fs.writeFileSync(configPath, JSON.stringify(config, null, 2), "utf-8")
+        // Write a temp file and rename it, so a crash mid-write can't leave
+        // a truncated file
+        const tempPath = `${configPath}.tmp`
+        fs.writeFileSync(tempPath, JSON.stringify(config, null, 2), "utf-8")
+        fs.renameSync(tempPath, configPath)
     } catch (error) {
         console.error("Failed to save proxy config:", error)
         throw error
@@ -47,6 +55,11 @@ export function saveProxyConfig(config: ProxyConfig): void {
 export function applyProxyToEnv(): void {
     const config = loadProxyConfig()
 
+    // No saved settings: keep proxy vars inherited from the system or .env
+    if (!config) {
+        return
+    }
+
     if (config.httpProxy) {
         process.env.HTTP_PROXY = config.httpProxy
         process.env.http_proxy = config.httpProxy

electron/main/window-manager.ts

@@ -3,6 +3,9 @@ import { app, BrowserWindow, screen } from "electron"
 
 let mainWindow: BrowserWindow | null = null
 
+// URL of the app server the main window loads
+let appUrl: string | null = null
+
 /**
  * Get the icon path based on platform
  * Note: electron-builder converts icon.png during packaging,
@@ -28,6 +31,7 @@ function getIconPath(): string | undefined {
  * Create the main application window
  */
 export function createWindow(serverUrl: string): BrowserWindow {
+    appUrl = serverUrl
     const { width, height } = screen.getPrimaryDisplay().workAreaSize
 
     mainWindow = new BrowserWindow({
@@ -56,7 +60,7 @@ export function createWindow(serverUrl: string): BrowserWindow {
     })
 
     // Open DevTools in development
-    if (process.env.NODE_ENV === "development") {
+    if (!app.isPackaged) {
         mainWindow.webContents.openDevTools()
     }
 
@@ -93,3 +97,39 @@ export function createWindow(serverUrl: string): BrowserWindow {
 export function getMainWindow(): BrowserWindow | null {
     return mainWindow
 }
+
+/**
+ * Get the app server URL the main window loads
+ */
+export function getAppUrl(): string | null {
+    return appUrl
+}
+
+/**
+ * Point the main window at the restarted app server (it can come up on a
+ * different port). On the same port the page fetches the new preset's
+ * server models instead of sending the old preset's choice; it is not
+ * reloaded, which would drop unsent attachments.
+ */
+export function setAppUrl(url: string): void {
+    if (url === appUrl) {
+        mainWindow?.webContents.send("server-restarted")
+        return
+    }
+    appUrl = url
+    mainWindow?.loadURL(url)
+}
+
+/**
+ * Check if a URL belongs to the app server (same origin)
+ */
+export function isAppUrl(url: string | undefined): boolean {
+    if (!url || !appUrl) {
+        return false
+    }
+    try {
+        return new URL(url).origin === new URL(appUrl).origin
+    } catch {
+        return false
+    }
+}

electron/preload/index.ts

@@ -18,10 +18,6 @@ contextBridge.exposeInMainWorld("electronAPI", {
     maximize: () => ipcRenderer.send("window-maximize"),
     close: () => ipcRenderer.send("window-close"),
 
-    // File operations
-    openFile: () => ipcRenderer.invoke("dialog-open-file"),
-    saveFile: (data: string) => ipcRenderer.invoke("dialog-save-file", data),
-
     // Proxy settings
     getProxy: () => ipcRenderer.invoke("get-proxy"),
     setProxy: (config: { httpProxy?: string; httpsProxy?: string }) =>
@@ -31,4 +27,18 @@ contextBridge.exposeInMainWorld("electronAPI", {
     getUserLocale: () => ipcRenderer.invoke("get-user-locale"),
     setUserLocale: (locale: string) =>
         ipcRenderer.invoke("set-user-locale", locale),
+
+    // A chat was saved, or the page loaded with this many chats: the next
+    // launch opens the port where the chats are
+    chatSaved: () => ipcRenderer.invoke("chat-saved"),
+    chatsLoaded: (count: number) => ipcRenderer.invoke("chats-loaded", count),
+
+    // The server restarted on the same port (another preset)
+    onServerRestarted: (callback: () => void) => {
+        const listener = () => callback()
+        ipcRenderer.on("server-restarted", listener)
+        return () => {
+            ipcRenderer.removeListener("server-restarted", listener)
+        }
+    },
 })

electron/settings/settings.js

@@ -213,6 +213,9 @@ async function savePreset() {
         }
     })
 
+    // closeModal() clears editingPresetId, so remember it for the toast
+    const isEdit = Boolean(editingPresetId)
+
     try {
         saveBtn.disabled = true
         saveBtn.innerHTML = '<span class="loading"></span>'
@@ -220,10 +223,7 @@ async function savePreset() {
         await window.settingsAPI.savePreset(preset)
         await loadPresets()
         closeModal()
-        showToast(
-            editingPresetId ? "Preset updated" : "Preset created",
-            "success",
-        )
+        showToast(isEdit ? "Preset updated" : "Preset created", "success")
     } catch (error) {
         console.error("Failed to save preset:", error)
         showToast("Failed to save preset", "error")
@@ -265,8 +265,6 @@ async function applyPreset(id) {
 
         const result = await window.settingsAPI.applyPreset(id)
         if (result.success) {
-            currentPresetId = id
-            renderPresets()
             showToast("Preset applied, server restarting...", "success")
         } else {
             showToast(result.error || "Failed to apply preset", "error")
@@ -274,6 +272,9 @@ async function applyPreset(id) {
     } catch (error) {
         console.error("Failed to apply preset:", error)
         showToast("Failed to apply preset", "error")
+    } finally {
+        // Reload to show the active preset and reset the Apply button
+        await loadPresets()
     }
 }
 

env.example

@@ -1,6 +1,6 @@
 # AI Provider Configuration
 # AI_PROVIDER: Which provider to use
-# Options: bedrock, openai, anthropic, google, vertexai, azure, ollama, openrouter, aihubmix, deepseek, siliconflow, gateway, novita
+# Options: bedrock, openai, anthropic, google, vertexai, azure, ollama, openrouter, aihubmix, deepseek, siliconflow, sglang, gateway, edgeone, doubao, modelscope, glm, qwen, qiniu, kimi, minimax, novita, mimo, atlascloud
 # Default: bedrock
 AI_PROVIDER=bedrock
 
@@ -11,6 +11,12 @@ AI_PROVIDER=bedrock
 # Example: AI_MODEL=doubao-seed-1-8-251215,doubao-seed-1-6-flash,doubao-seed-1-6-pro
 AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 
+# Output limit, all providers (default: 64000). Shared by reasoning and the diagram XML,
+# so a thinking model can spend it all before the tool call. Users can lower it in Settings,
+# and raise it only when they use their own API key, so this also caps cost on server keys.
+# If a model's own ceiling is lower, the request is retried with that ceiling automatically.
+# MAX_OUTPUT_TOKENS=64000
+
 # AWS Bedrock Configuration
 # AWS_REGION=us-east-1
 # AWS_ACCESS_KEY_ID=your-access-key-id
@@ -38,7 +44,6 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 # Google Generative AI Configuration
 # GOOGLE_GENERATIVE_AI_API_KEY=...
 # GOOGLE_BASE_URL=https://generativelanguage.googleapis.com/v1beta  # Optional: Custom endpoint
-# GOOGLE_CANDIDATE_COUNT=1                   # Optional: Number of candidates to generate
 # GOOGLE_TOP_K=40                            # Optional: Top K sampling parameter
 # GOOGLE_TOP_P=0.95                          # Optional: Nucleus sampling parameter
 # Note: Gemini 2.5/3 models automatically enable reasoning display (includeThoughts: true)
@@ -65,7 +70,7 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 # AZURE_REASONING_SUMMARY=detailed
 
 # Ollama Configuration (Local or Cloud)
-# OLLAMA_BASE_URL=https://ollama.com/api  # Optional, defaults to Ollama Cloud
+# OLLAMA_BASE_URL=https://ollama.com/api  # Optional: Ollama Cloud; defaults to local Ollama (http://127.0.0.1:11434/api)
 # OLLAMA_API_KEY=your-ollama-cloud-api-key    # Optional: For Ollama Cloud or authenticated remote instances
 # OLLAMA_ENABLE_THINKING=true                 # Optional: Enable thinking for models that support it (e.g., qwen3)
 
@@ -118,9 +123,14 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 
 # Temperature (Optional)
 # Controls randomness in AI responses. Lower = more deterministic.
-# Leave unset for models that don't support temperature (e.g., GPT-5.1 reasoning models)
+# Leave unset for models that don't support temperature (e.g., GPT-5.1 reasoning models).
+# Claude 4.7 and later reject it; the request is then retried without it.
 # TEMPERATURE=0
 
+# Debug Logging (Optional)
+# Log the structure of the messages each chat request sends to the model
+# DEBUG_LLM_PAYLOAD=true
+
 # Access Control (Optional)
 # ACCESS_CODE_LIST=your-secret-code,another-code
 
@@ -153,11 +163,22 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 # Set to "false" to block private IPs, localhost, and internal hostnames
 # ALLOW_PRIVATE_URLS=false
 
+# Behind a CDN such as Cloudflare (Optional)
+# The daily quota is counted per IP. By default the IP is the first
+# X-Forwarded-For entry, which visitors can set to anything. Name the header
+# your CDN fills with the visitor's real IP instead:
+# CLIENT_IP_HEADER=cf-connecting-ip
+# Then have the CDN add an X-Origin-Secret header with this value to every
+# request. API calls without it get 403, so nobody can skip the CDN and fake
+# the IP header above.
+# ORIGIN_SECRET=a-long-random-string
+
 # Self-hosted deployment (Optional)
 # Self-hosted users may implement custom quota-management solutions,
 # which triggers the client UI to display messages suggesting self-hosting or sponsorship.
 # This switch allows self-hosted users to provide custom messages in response to a 429 code,
 # in messageTokenSelfHosted, messageApiSelfHosted, and tipSelfHosted translation strings.
+# It also turns off the MCP server advertisement in the chat examples.
 # NEXT_PUBLIC_SELFHOSTED=true
 
 # Minimax Configuration (Optional)

hooks/use-diagram-tool-handlers.ts

@@ -1,18 +1,20 @@
 import type { MutableRefObject } from "react"
-import { useRef } from "react"
 import type { DiagramOperation } from "@/components/chat/types"
 import type {
     ValidationState,
     ValidationStatus,
 } from "@/components/chat/ValidationCard"
-import type { Operation } from "@/lib/diagram-engine"
-import { restructureDiagram } from "@/lib/diagram-engine"
 import type { ValidationResult } from "@/lib/diagram-validator"
 import { formatValidationFeedback } from "@/lib/diagram-validator"
-import { isMxCellXmlComplete, isRealDiagram, wrapWithMxFile } from "@/lib/utils"
+import { isMxCellXmlComplete } from "@/lib/utils"
+import { editDiagram } from "@/packages/mcp-server/src/edit-diagram.ts"
+import { prepareNewDiagram } from "@/packages/mcp-server/src/new-diagram.ts"
 
 const DEBUG = process.env.NODE_ENV === "development"
 
+// display_diagram replaces the document with this one page
+const NEW_PAGE = { pageId: "page-1", pageName: "Page-1" }
+
 interface ToolCall {
     toolCallId: string
     toolName: string
@@ -50,6 +52,10 @@ type ValidateDiagramFn = (
 interface UseDiagramToolHandlersParams {
     partialXmlRef: MutableRefObject<string>
     editDiagramOriginalXmlRef: MutableRefObject<Map<string, string>>
+    // Tool calls the streaming preview must leave alone (shared with it)
+    processedToolCallsRef: MutableRefObject<Set<string>>
+    // Failed VLM validations in the current user turn (reset on each user message)
+    validationRetryCountRef: MutableRefObject<number>
     chartXMLRef: MutableRefObject<string>
     onDisplayChart: (xml: string, skipValidation?: boolean) => string | null
     onFetchChart: (saveToHistory?: boolean) => Promise<string>
@@ -58,6 +64,11 @@ interface UseDiagramToolHandlersParams {
     validateDiagram?: ValidateDiagramFn
     enableVlmValidation?: boolean
     sessionId?: string
+    // Called when a screenshot check begins; the function it returns
+    // tells whether the user pressed Stop in this turn, also after the next
+    // message was sent. A check that has not started then is skipped (one
+    // already running is cancelled by the caller).
+    watchStop?: () => () => boolean
     onValidationStateChange?: (
         toolCallId: string,
         state: ValidationState,
@@ -66,8 +77,7 @@ interface UseDiagramToolHandlersParams {
 
 /**
  * Hook that creates the onToolCall handler for diagram-related tools.
- * Handles edit_diagram and restructure_diagram, plus the cached-XML replay that arrives
- * as display_diagram.
+ * Handles display_diagram, edit_diagram, and append_diagram tools.
  *
  * Note: addToolOutput is passed at call time (not hook init) because
  * it comes from useChat which creates a circular dependency.
@@ -75,6 +85,8 @@ interface UseDiagramToolHandlersParams {
 export function useDiagramToolHandlers({
     partialXmlRef,
     editDiagramOriginalXmlRef,
+    processedToolCallsRef,
+    validationRetryCountRef,
     chartXMLRef,
     onDisplayChart,
     onFetchChart,
@@ -83,11 +95,9 @@ export function useDiagramToolHandlers({
     validateDiagram,
     enableVlmValidation = true,
     sessionId,
+    watchStop,
     onValidationStateChange,
 }: UseDiagramToolHandlersParams) {
-    // Track validation retry count per tool call
-    const validationRetryCountRef = useRef<Map<string, number>>(new Map())
-
     // Helper to update validation state
     const updateValidationState = (
         toolCallId: string,
@@ -117,57 +127,312 @@ export function useDiagramToolHandlers({
             )
         }
 
+        processedToolCallsRef.current.add(toolCall.toolCallId)
+        // Only display_diagram, edit_diagram and a completing append_diagram
+        // put their result on the canvas. Other tools (get_shape_library,
+        // which the server runs, still arrives here) leave the stored
+        // originals for the preview code to undo.
         if (toolCall.toolName === "display_diagram") {
-            await handleDisplayDiagram(toolCall, addToolOutput)
+            await handleDisplayDiagram(toolCall, addToolOutput, takeOriginals())
         } else if (toolCall.toolName === "edit_diagram") {
-            await handleEditDiagram(toolCall, addToolOutput)
-        } else if (toolCall.toolName === "restructure_diagram") {
-            await handleRestructureDiagram(toolCall, addToolOutput)
+            await handleEditDiagram(toolCall, addToolOutput, takeOriginals())
+        } else if (toolCall.toolName === "append_diagram") {
+            handleAppendDiagram(toolCall, addToolOutput)
         }
     }
 
-    // Replays a cached XML answer onto the canvas. The model can no longer call this tool —
-    // it only arrives from the server's cache-hit path (see createCachedStreamResponse), which
-    // speaks the same wire format. So there is no truncation to continue and no model to send
-    // errors back to: load it, or report that it did not load.
+    // Stored originals belong to previews not handled yet: this call's, and
+    // those of earlier calls with invalid input, which never get to the
+    // handler. The first is the diagram before all of them. A call that
+    // draws its result replaces those previews, so the preview code must
+    // neither draw them again nor undo them later. Returns that first one.
+    const takeOriginals = (): string | undefined => {
+        const [originalXml] = editDiagramOriginalXmlRef.current.values()
+        for (const id of editDiagramOriginalXmlRef.current.keys()) {
+            processedToolCallsRef.current.add(id)
+        }
+        editDiagramOriginalXmlRef.current.clear()
+        return originalXml
+    }
+
+    // originalXml: the diagram before the streamed previews, if any were drawn
     const handleDisplayDiagram = async (
         toolCall: ToolCall,
         addToolOutput: AddToolOutputFn,
+        originalXml: string | undefined,
     ) => {
         const { xml } = toolCall.input as { xml: string }
-        const validationError = onDisplayChart(wrapWithMxFile(xml))
-        if (validationError) {
-            console.warn("[display_diagram] Validation error:", validationError)
+
+        // DEBUG: Log raw input to diagnose false truncation detection
+        if (DEBUG) {
+            console.log(
+                "[display_diagram] XML ending (last 100 chars):",
+                xml.slice(-100),
+            )
+            console.log("[display_diagram] XML length:", xml.length)
+        }
+
+        // Check if XML is truncated (incomplete mxCell indicates truncated output)
+        const isTruncated = !isMxCellXmlComplete(xml)
+        if (DEBUG) {
+            console.log("[display_diagram] isTruncated:", isTruncated)
+        }
+
+        if (isTruncated) {
+            // Store the partial XML for continuation via append_diagram
+            partialXmlRef.current = xml
+
+            // Tell LLM to use append_diagram to continue
+            const partialEnding = partialXmlRef.current.slice(-500)
             addToolOutput({
                 tool: "display_diagram",
                 toolCallId: toolCall.toolCallId,
                 state: "output-error",
-                errorText: validationError,
+                errorText: `Output was truncated due to length limits. Use the append_diagram tool to continue.
+
+Your output ended with:
+\`\`\`
+${partialEnding}
+\`\`\`
+
+NEXT STEP: Call append_diagram with the continuation XML.
+- Do NOT include wrapper tags or root cells (id="0", id="1")
+- Start from EXACTLY where you stopped
+- Complete all remaining mxCell elements`,
             })
             return
         }
-        addToolOutput({
-            tool: "display_diagram",
-            toolCallId: toolCall.toolCallId,
-            output: "Successfully displayed the diagram.",
-        })
+
+        // Complete XML received - use it directly
+        // (continuation is now handled via append_diagram tool)
+        const finalXml = xml
+        partialXmlRef.current = "" // Reset any partial from previous truncation
+
+        // Wrap, validate and auto-fix the model's XML like the MCP server's
+        // create_new_diagram, then load it
+        const prepared = prepareNewDiagram(finalXml, NEW_PAGE)
+        const validationError = prepared.ok
+            ? onDisplayChart(prepared.xml, true)
+            : prepared.error
+
+        if (validationError) {
+            console.warn("[display_diagram] Validation error:", validationError)
+            // Undo the streamed preview, as a failed edit does: the canvas
+            // keeps the diagram from before this failed call
+            if (originalXml) onDisplayChart(originalXml, true)
+            // Return error to model - sendAutomaticallyWhen will trigger retry
+            if (DEBUG) {
+                console.log(
+                    "[display_diagram] Adding tool output with state: output-error",
+                )
+            }
+            addToolOutput({
+                tool: "display_diagram",
+                toolCallId: toolCall.toolCallId,
+                state: "output-error",
+                errorText: `${validationError}
+
+Please fix the XML issues and call display_diagram again with corrected XML.
+
+Your failed XML:
+\`\`\`xml
+${finalXml}
+\`\`\``,
+            })
+        } else {
+            // Success - diagram will be rendered by chat-message-display
+            if (DEBUG) {
+                console.log(
+                    "[display_diagram] Success! Checking if VLM validation is enabled...",
+                )
+            }
+
+            // VLM validation after successful display
+            if (
+                enableVlmValidation &&
+                captureValidationPng &&
+                validateDiagram &&
+                // At most this many checks per user turn, passed or not
+                validationRetryCountRef.current < MAX_VALIDATION_RETRIES
+            ) {
+                let capturedPngData: string | null = null
+                const stopped = watchStop?.()
+                try {
+                    // Notify UI that we're starting capture
+                    updateValidationState(toolCall.toolCallId, "capturing")
+
+                    // Small delay (100ms) to allow diagram rendering to complete before capture.
+                    // This is a best-effort heuristic and may need adjustment for complex diagrams or slower devices.
+                    await new Promise((resolve) => setTimeout(resolve, 100))
+
+                    capturedPngData = await captureValidationPng()
+                    // Stopped while the screenshot was taken: no check. The
+                    // chat waits for this handler, so it must end now.
+                    if (stopped?.()) {
+                        updateValidationState(toolCall.toolCallId, "skipped")
+                    } else if (capturedPngData) {
+                        if (DEBUG) {
+                            console.log(
+                                "[display_diagram] Captured PNG for validation",
+                            )
+                        }
+
+                        // Each retry is a new tool call, so count attempts
+                        // per user turn (the chat resets it when the user sends)
+                        const attempt = validationRetryCountRef.current + 1
+                        validationRetryCountRef.current = attempt
+
+                        // Notify UI that we're validating (include the image)
+                        updateValidationState(
+                            toolCall.toolCallId,
+                            "validating",
+                            {
+                                attempt,
+                                maxAttempts: MAX_VALIDATION_RETRIES,
+                                imageData: capturedPngData,
+                            },
+                        )
+
+                        const result = await validateDiagram(
+                            capturedPngData,
+                            sessionId,
+                        )
+
+                        if (!result.valid) {
+                            if (attempt < MAX_VALIDATION_RETRIES) {
+                                const feedback =
+                                    formatValidationFeedback(result)
+                                if (DEBUG) {
+                                    console.log(
+                                        `[display_diagram] Validation failed (attempt ${attempt}/${MAX_VALIDATION_RETRIES}):`,
+                                        result.issues,
+                                    )
+                                }
+
+                                // Notify UI of validation failure (include the image)
+                                updateValidationState(
+                                    toolCall.toolCallId,
+                                    "failed",
+                                    {
+                                        attempt,
+                                        maxAttempts: MAX_VALIDATION_RETRIES,
+                                        result,
+                                        imageData: capturedPngData,
+                                    },
+                                )
+
+                                addToolOutput({
+                                    tool: "display_diagram",
+                                    toolCallId: toolCall.toolCallId,
+                                    state: "output-error",
+                                    errorText: `[Validation attempt ${attempt}/${MAX_VALIDATION_RETRIES}]\n${feedback}`,
+                                })
+                                return
+                            } else {
+                                // Last attempt - accept the diagram with warning
+                                if (DEBUG) {
+                                    console.log(
+                                        "[display_diagram] Max validation retries reached, accepting diagram",
+                                    )
+                                }
+                                // Notify UI that we're accepting with issues (include the image)
+                                updateValidationState(
+                                    toolCall.toolCallId,
+                                    "skipped",
+                                    { result, imageData: capturedPngData },
+                                )
+
+                                addToolOutput({
+                                    tool: "display_diagram",
+                                    toolCallId: toolCall.toolCallId,
+                                    output: "Diagram displayed (validation issues noted but max retries reached).",
+                                })
+                                return
+                            }
+                        } else {
+                            if (DEBUG) {
+                                console.log(
+                                    "[display_diagram] Validation passed!",
+                                )
+                            }
+
+                            // Notify UI of success (include the image)
+                            // Use "success_with_warnings" if valid but has issues
+                            const hasWarnings = result.issues.length > 0
+                            updateValidationState(
+                                toolCall.toolCallId,
+                                hasWarnings
+                                    ? "success_with_warnings"
+                                    : "success",
+                                { result, imageData: capturedPngData },
+                            )
+                        }
+                    } else {
+                        // PNG capture failed - skip validation
+                        updateValidationState(toolCall.toolCallId, "skipped")
+                    }
+                } catch (error) {
+                    // Cancelled by Stop: the diagram stays, unchecked
+                    if ((error as Error)?.name === "AbortError") {
+                        updateValidationState(toolCall.toolCallId, "skipped")
+                        addToolOutput({
+                            tool: "display_diagram",
+                            toolCallId: toolCall.toolCallId,
+                            output: "Successfully displayed the diagram.",
+                        })
+                        return
+                    }
+                    // VLM validation error - log but don't block the user
+                    console.warn(
+                        "[display_diagram] VLM validation error:",
+                        error,
+                    )
+                    updateValidationState(toolCall.toolCallId, "error", {
+                        error:
+                            error instanceof Error
+                                ? error.message
+                                : "Validation failed",
+                        imageData: capturedPngData || undefined,
+                    })
+                }
+            }
+
+            if (DEBUG) {
+                console.log(
+                    "[display_diagram] Adding tool output with state: output-available",
+                )
+            }
+            addToolOutput({
+                tool: "display_diagram",
+                toolCallId: toolCall.toolCallId,
+                output: "Successfully displayed the diagram.",
+            })
+            if (DEBUG) {
+                console.log(
+                    "[display_diagram] Tool output added. Diagram should be visible now.",
+                )
+            }
+        }
     }
 
+    // originalXml: the diagram before the streamed previews, if any were drawn.
+    // Operations apply to it, the same base XML that streaming used.
     const handleEditDiagram = async (
         toolCall: ToolCall,
         addToolOutput: AddToolOutputFn,
+        originalXml: string | undefined,
     ) => {
         const { operations } = toolCall.input as {
             operations: DiagramOperation[]
         }
 
         let currentXml = ""
+        // On failure, undo the streaming preview so the canvas matches the XML
+        // reported back to the model
+        const restoreOriginal = () => {
+            if (originalXml) onDisplayChart(originalXml, true)
+        }
         try {
-            // Use the original XML captured during streaming (shared with chat-message-display)
-            // This ensures we apply operations to the same base XML that streaming used
-            const originalXml = editDiagramOriginalXmlRef.current.get(
-                toolCall.toolCallId,
-            )
             if (originalXml) {
                 currentXml = originalXml
             } else {
@@ -181,26 +446,19 @@ export function useDiagramToolHandlers({
                 }
             }
 
-            const { applyDiagramOperations } = await import("@/lib/utils")
-            const { result: editedXml, errors } = applyDiagramOperations(
-                currentXml,
-                operations,
-            )
-
-            // Check for operation errors
-            if (errors.length > 0) {
-                const errorMessages = errors
-                    .map(
-                        (e) =>
-                            `- ${e.type} on cell_id="${e.cellId}": ${e.message}`,
-                    )
-                    .join("\n")
-
+            // All or nothing, checked like the MCP server's edit_diagram.
+            // The model sees the first page, so edits target it.
+            const outcome = editDiagram(currentXml, operations, {})
+            if (!outcome.ok) {
+                const reason = outcome.pageError
+                    ? outcome.errors[0]
+                    : `No changes were made because ${outcome.errors.length} operation(s) failed:\n${outcome.errors.map((e) => `- ${e}`).join("\n")}`
+                restoreOriginal()
                 addToolOutput({
                     tool: "edit_diagram",
                     toolCallId: toolCall.toolCallId,
                     state: "output-error",
-                    errorText: `Some operations failed:\n${errorMessages}
+                    errorText: `${reason}
 
 Current diagram XML:
 \`\`\`xml
@@ -209,49 +467,23 @@ ${currentXml}
 
 Please check the cell IDs and retry.`,
                 })
-                // Clean up the shared original XML ref
-                editDiagramOriginalXmlRef.current.delete(toolCall.toolCallId)
                 return
             }
 
-            // loadDiagram validates and returns error if invalid
-            const validationError = onDisplayChart(editedXml)
-            if (validationError) {
-                console.warn(
-                    "[edit_diagram] Validation error:",
-                    validationError,
-                )
-                addToolOutput({
-                    tool: "edit_diagram",
-                    toolCallId: toolCall.toolCallId,
-                    state: "output-error",
-                    errorText: `Edit produced invalid XML: ${validationError}
-
-Current diagram XML:
-\`\`\`xml
-${currentXml}
-\`\`\`
-
-Please fix the operations to avoid structural issues.`,
-                })
-                // Clean up the shared original XML ref
-                editDiagramOriginalXmlRef.current.delete(toolCall.toolCallId)
-                return
-            }
+            onDisplayChart(outcome.xml, true)
             onExport()
             addToolOutput({
                 tool: "edit_diagram",
                 toolCallId: toolCall.toolCallId,
-                output: `Successfully applied ${operations.length} operation(s) to the diagram.`,
+                output: `Successfully applied ${outcome.applied} operation(s) to the diagram.`,
             })
-            // Clean up the shared original XML ref
-            editDiagramOriginalXmlRef.current.delete(toolCall.toolCallId)
         } catch (error) {
             console.error("[edit_diagram] Failed:", error)
 
             const errorMessage =
                 error instanceof Error ? error.message : String(error)
 
+            restoreOriginal()
             addToolOutput({
                 tool: "edit_diagram",
                 toolCallId: toolCall.toolCallId,
@@ -263,77 +495,116 @@ Current diagram XML:
 ${currentXml || "No XML available"}
 \`\`\`
 
-Please check cell IDs and retry, or rebuild with restructure_diagram.`,
+Please check cell IDs and retry, or use display_diagram to regenerate.`,
             })
-            // Clean up the shared original XML ref even on error
-            editDiagramOriginalXmlRef.current.delete(toolCall.toolCallId)
         }
     }
 
-    /**
-     * Structural editing. The model sends operations against the tree; the engine
-     * re-derives that tree from whatever is on the canvas right now — including anything
-     * the user moved or recoloured by hand — applies the operations, recomputes every
-     * coordinate, and returns new XML.
-     *
-     * Nothing about the tree is stored between calls, so there is no second copy of the
-     * state to drift out of sync with the canvas.
-     */
-    const handleRestructureDiagram = async (
+    const handleAppendDiagram = (
         toolCall: ToolCall,
         addToolOutput: AddToolOutputFn,
     ) => {
-        const { operations } = toolCall.input as { operations: Operation[] }
+        const { xml } = toolCall.input as { xml: string }
 
-        // Read the live canvas, not the last thing we generated: the user may have
-        // edited it since.
-        let currentXml = ""
-        try {
-            currentXml = await onFetchChart(false)
-        } catch {
-            currentXml = chartXMLRef.current ?? ""
-        }
-        if (!isRealDiagram(currentXml)) currentXml = ""
-
-        const result = restructureDiagram(currentXml, operations)
-
-        if (result.errors.length > 0 || !result.xml) {
+        // Nothing to continue: loading the fragment alone would replace the whole diagram
+        if (!partialXmlRef.current) {
             addToolOutput({
-                tool: "restructure_diagram",
+                tool: "append_diagram",
                 toolCallId: toolCall.toolCallId,
                 state: "output-error",
-                errorText: `Could not apply the operations:
-${result.errors.map((e) => `- ${e}`).join("\n")}
+                errorText: `ERROR: There is no truncated diagram to continue, so append_diagram cannot be used now.
 
-Structure as it stands:
-${result.outline}
-
-Fix the operations and call restructure_diagram again.`,
+Use display_diagram to create the complete diagram, or edit_diagram to change the current one.`,
             })
             return
         }
 
-        const loadError = onDisplayChart(result.xml)
-        if (loadError) {
+        // Detect if LLM incorrectly started fresh instead of continuing
+        // LLM should only output bare mxCells now, so wrapper tags indicate error
+        const trimmed = xml.trim()
+        const isFreshStart =
+            trimmed.startsWith("<mxGraphModel") ||
+            trimmed.startsWith("<root") ||
+            trimmed.startsWith("<mxfile") ||
+            trimmed.startsWith('<mxCell id="0"') ||
+            trimmed.startsWith('<mxCell id="1"')
+
+        if (isFreshStart) {
             addToolOutput({
-                tool: "restructure_diagram",
+                tool: "append_diagram",
                 toolCallId: toolCall.toolCallId,
                 state: "output-error",
-                errorText: `The diagram was built but draw.io rejected it: ${loadError}`,
+                errorText: `ERROR: You started fresh with wrapper tags. Do NOT include wrapper tags or root cells (id="0", id="1").
+
+Continue from EXACTLY where the partial ended:
+\`\`\`
+${partialXmlRef.current.slice(-500)}
+\`\`\`
+
+Start your continuation with the NEXT character after where it stopped.`,
             })
             return
         }
 
-        // Report the outline rather than the XML: it is what the model needs to name ids
-        // in the next call, at a fraction of the tokens.
-        const notes = result.warnings.length
-            ? `\n\nNotes:\n${result.warnings.map((w) => `- ${w}`).join("\n")}`
-            : ""
-        addToolOutput({
-            tool: "restructure_diagram",
-            toolCallId: toolCall.toolCallId,
-            output: `Diagram updated.\n\n${result.outline}${notes}`,
-        })
+        // Append to accumulated XML
+        partialXmlRef.current += xml
+
+        // Check if XML is now complete (last mxCell is complete)
+        const isComplete = isMxCellXmlComplete(partialXmlRef.current)
+
+        if (isComplete) {
+            // Wrap and display the complete diagram
+            const finalXml = partialXmlRef.current
+            partialXmlRef.current = "" // Reset
+
+            const prepared = prepareNewDiagram(finalXml, NEW_PAGE)
+            // It draws now: it takes the stored originals, as display_diagram
+            const originalXml = prepared.ok ? takeOriginals() : undefined
+            const validationError = prepared.ok
+                ? onDisplayChart(prepared.xml, true)
+                : prepared.error
+
+            if (validationError) {
+                // Loading failed: back to the diagram before the previews
+                if (prepared.ok && originalXml) {
+                    onDisplayChart(originalXml, true)
+                }
+                addToolOutput({
+                    tool: "append_diagram",
+                    toolCallId: toolCall.toolCallId,
+                    state: "output-error",
+                    errorText: `Validation error after assembly: ${validationError}
+
+Assembled XML:
+\`\`\`xml
+${finalXml.substring(0, 2000)}...
+\`\`\`
+
+Please use display_diagram with corrected XML.`,
+                })
+            } else {
+                addToolOutput({
+                    tool: "append_diagram",
+                    toolCallId: toolCall.toolCallId,
+                    output: "Diagram assembly complete and displayed successfully.",
+                })
+            }
+        } else {
+            // Still incomplete - signal to continue
+            addToolOutput({
+                tool: "append_diagram",
+                toolCallId: toolCall.toolCallId,
+                state: "output-error",
+                errorText: `XML still incomplete (mxCell not closed). Call append_diagram again to continue.
+
+Current ending:
+\`\`\`
+${partialXmlRef.current.slice(-500)}
+\`\`\`
+
+Continue from EXACTLY where you stopped.`,
+            })
+        }
     }
 
     return { handleToolCall }

hooks/use-model-config.ts

@@ -13,6 +13,7 @@ import {
     flattenModels,
     type ModelConfig,
     type MultiModelConfig,
+    PROVIDER_INFO,
     type ProviderConfig,
     type ProviderName,
 } from "@/lib/types/model-config"
@@ -63,6 +64,28 @@ function migrateOldConfig(): MultiModelConfig | null {
     return config
 }
 
+const isKnownProvider = (p: { provider: string }) =>
+    Object.hasOwn(PROVIDER_INFO, p.provider)
+
+/**
+ * The stored config without providers this version does not know (saved
+ * by another version, or edited by hand): they would break every list of
+ * models. They stay in storage (saveConfig keeps them). Throws on bad JSON.
+ */
+function parseStoredConfig(stored: string): MultiModelConfig {
+    const config = JSON.parse(stored) as MultiModelConfig
+    const known = config.providers.filter(isKnownProvider)
+    if (known.length < config.providers.length) {
+        console.warn(
+            "Skipped saved providers this version does not know:",
+            config.providers
+                .filter((p) => !isKnownProvider(p))
+                .map((p) => p.provider),
+        )
+    }
+    return { ...config, providers: known }
+}
+
 /**
  * Load config from localStorage
  */
@@ -73,7 +96,7 @@ function loadConfig(): MultiModelConfig {
     const stored = localStorage.getItem(STORAGE_KEYS.modelConfigs)
     if (stored) {
         try {
-            return JSON.parse(stored) as MultiModelConfig
+            return parseStoredConfig(stored)
         } catch {
             console.error("Failed to parse model config")
         }
@@ -98,7 +121,35 @@ function loadConfig(): MultiModelConfig {
  */
 function saveConfig(config: MultiModelConfig): void {
     if (typeof window === "undefined") return
-    localStorage.setItem(STORAGE_KEYS.modelConfigs, JSON.stringify(config))
+    // Providers this version does not know are not in config: keep them,
+    // with their keys, for the version that saved them
+    let unknown: MultiModelConfig["providers"] = []
+    try {
+        const stored = localStorage.getItem(STORAGE_KEYS.modelConfigs)
+        if (stored) {
+            unknown = (JSON.parse(stored) as MultiModelConfig).providers.filter(
+                (p) => !isKnownProvider(p),
+            )
+        }
+    } catch {
+        // Unreadable: nothing to keep
+    }
+    localStorage.setItem(
+        STORAGE_KEYS.modelConfigs,
+        JSON.stringify({
+            ...config,
+            providers: [...config.providers, ...unknown],
+        }),
+    )
+}
+
+/**
+ * Server model to fall back to: the one marked default, else the first one
+ */
+function defaultServerModelId(
+    serverModels: FlattenedServerModel[],
+): string | undefined {
+    return (serverModels.find((m) => m.isDefault) ?? serverModels[0])?.id
 }
 
 export interface UseModelConfigReturn {
@@ -144,10 +195,25 @@ export function useModelConfig(): UseModelConfigReturn {
         setIsLoaded(true)
     }, [])
 
-    // Load server models on mount (if any)
+    // Pick up config changes saved by other tabs, so this tab neither shows a
+    // stale model nor overwrites their changes on its next save
+    useEffect(() => {
+        const handleStorage = (e: StorageEvent) => {
+            if (e.key === STORAGE_KEYS.modelConfigs) setConfig(loadConfig())
+        }
+        window.addEventListener("storage", handleStorage)
+        return () => window.removeEventListener("storage", handleStorage)
+    }, [])
+
+    // Load server models on mount (if any), and again when the desktop app
+    // restarted its server for another preset
     useEffect(() => {
         if (typeof window === "undefined") return
+        loadServerModels()
+        return window.electronAPI?.onServerRestarted?.(loadServerModels)
+    }, [])
 
+    function loadServerModels() {
         fetch(getApiEndpoint("/api/server-models"))
             .then((res) => {
                 if (!res.ok) {
@@ -165,24 +231,37 @@ export function useModelConfig(): UseModelConfigReturn {
                 setServerModels(raw)
                 setServerLoaded(true)
 
-                // Auto-select default server model if no model is currently selected
+                // Auto-select the default server model if no model is selected,
+                // or if the saved server model is gone (renamed or removed)
                 setConfig((prev) => {
-                    if (!prev.selectedModelId && raw.length > 0) {
-                        const defaultModel = raw.find((m) => m.isDefault)
-                        if (defaultModel) {
-                            return { ...prev, selectedModelId: defaultModel.id }
-                        }
-                        // If no default marked, use first server model
-                        return { ...prev, selectedModelId: raw[0].id }
-                    }
-                    return prev
+                    const id = prev.selectedModelId
+                    const isStale =
+                        id?.startsWith("server:") &&
+                        !raw.some((m) => m.id === id)
+                    if (id && !isStale) return prev
+                    // Saved before non-ASCII characters in provider names
+                    // got into the id: they were dropped from it
+                    const renamed = raw.filter(
+                        (m) =>
+                            `server:${m.providerLabel
+                                .toLowerCase()
+                                .replace(/[^a-z0-9]+/g, "-")
+                                .replace(/^-|-$/g, "")}:${m.modelId}` === id,
+                    )
+                    const fallback =
+                        renamed.length === 1
+                            ? renamed[0].id
+                            : defaultServerModelId(raw)
+                    return fallback === id
+                        ? prev
+                        : { ...prev, selectedModelId: fallback }
                 })
             })
             .catch((error) => {
                 console.error("Error while loading server models:", error)
                 setServerLoaded(true)
             })
-    }, [])
+    }
 
     // Save config whenever it changes (after initial load)
     useEffect(() => {
@@ -260,24 +339,31 @@ export function useModelConfig(): UseModelConfigReturn {
         [],
     )
 
-    const deleteProvider = useCallback((providerId: string) => {
-        setConfig((prev) => {
-            const provider = prev.providers.find((p) => p.id === providerId)
-            const modelIds = provider?.models.map((m) => m.id) || []
+    const deleteProvider = useCallback(
+        (providerId: string) => {
+            setConfig((prev) => {
+                const provider = prev.providers.find((p) => p.id === providerId)
+                const modelIds = provider?.models.map((m) => m.id) || []
 
-            // Clear selected model if it belongs to deleted provider
-            const newSelectedId =
-                prev.selectedModelId && modelIds.includes(prev.selectedModelId)
-                    ? undefined
-                    : prev.selectedModelId
+                // Fall back to the default server model if the selected model
+                // belongs to the deleted provider
+                const newSelectedId =
+                    prev.selectedModelId &&
+                    modelIds.includes(prev.selectedModelId)
+                        ? defaultServerModelId(serverModels)
+                        : prev.selectedModelId
 
-            return {
-                ...prev,
-                providers: prev.providers.filter((p) => p.id !== providerId),
-                selectedModelId: newSelectedId,
-            }
-        })
-    }, [])
+                return {
+                    ...prev,
+                    providers: prev.providers.filter(
+                        (p) => p.id !== providerId,
+                    ),
+                    selectedModelId: newSelectedId,
+                }
+            })
+        },
+        [serverModels],
+    )
 
     const addModel = useCallback(
         (providerId: string, modelId: string): ModelConfig => {
@@ -334,14 +420,15 @@ export function useModelConfig(): UseModelConfigReturn {
                           }
                         : p,
                 ),
-                // Clear selected model if it was deleted
+                // Fall back to the default server model if the selected model
+                // was deleted
                 selectedModelId:
                     prev.selectedModelId === modelConfigId
-                        ? undefined
+                        ? defaultServerModelId(serverModels)
                         : prev.selectedModelId,
             }))
         },
-        [],
+        [serverModels],
     )
 
     const resetConfig = useCallback(() => {
@@ -428,7 +515,8 @@ export function getSelectedAIConfig(): {
 
     let config: MultiModelConfig
     try {
-        config = JSON.parse(stored)
+        // Unknown providers would break the model lookup below
+        config = parseStoredConfig(stored)
     } catch {
         return { ...empty, accessCode }
     }

hooks/use-session-manager.ts

@@ -1,6 +1,8 @@
 "use client"
 
 import { useCallback, useEffect, useRef, useState } from "react"
+import { toast } from "sonner"
+import { useDictionary } from "@/hooks/use-dictionary"
 import {
     type ChatSession,
     createEmptySession,
@@ -11,10 +13,12 @@ import {
     getSession,
     isIndexedDBAvailable,
     migrateFromLocalStorage,
+    readSessionCount,
     type SessionMetadata,
     type StoredMessage,
     saveSession,
 } from "@/lib/session-storage"
+import { STORAGE_KEYS } from "@/lib/storage"
 
 export interface SessionData {
     messages: StoredMessage[]
@@ -24,6 +28,13 @@ export interface SessionData {
     diagramHistory?: { svg: string; xml: string }[]
 }
 
+// Taken right before a save's data is read: the chat on screen then, and
+// the order of the reads
+export interface SaveTicket {
+    generation: number
+    seq: number
+}
+
 export interface UseSessionManagerReturn {
     // State
     sessions: SessionMetadata[]
@@ -35,13 +46,27 @@ export interface UseSessionManagerReturn {
     // Actions
     switchSession: (id: string) => Promise<SessionData | null>
     deleteSession: (id: string) => Promise<{ wasCurrentSession: boolean }>
-    // forSessionId: optional session ID to verify save targets correct session (prevents stale debounce writes)
+    // ticket: getSaveTicket() before the data was read (by default, now).
+    // The save is dropped if another chat is on screen when its turn comes,
+    // or if a copy of this chat read later was saved already.
+    // Resolves to false when the save failed (the user was told)
     saveCurrentSession: (
         data: SessionData,
-        forSessionId?: string | null,
-    ) => Promise<void>
+        ticket?: SaveTicket,
+    ) => Promise<boolean>
     refreshSessions: () => Promise<void>
     clearCurrentSession: () => void
+    getChatGeneration: () => number
+    getSaveTicket: () => SaveTicket
+}
+
+// Reading the session list loads every stored session in full, and window
+// focus also fires each time the user clicks back from the draw.io iframe
+const FOCUS_REFRESH_INTERVAL_MS = 30_000
+
+function notifySaveFailed(message: string) {
+    // Same id, so repeated failures update one toast instead of stacking
+    toast.error(message, { id: "session-save-failed", duration: 8000 })
 }
 
 interface UseSessionManagerOptions {
@@ -53,6 +78,7 @@ export function useSessionManager(
     options: UseSessionManagerOptions = {},
 ): UseSessionManagerReturn {
     const { initialSessionId } = options
+    const dict = useDictionary()
     const [sessions, setSessions] = useState<SessionMetadata[]>([])
     const [currentSessionId, setCurrentSessionId] = useState<string | null>(
         null,
@@ -66,6 +92,23 @@ export function useSessionManager(
     const isInitializedRef = useRef(false)
     // Sequence guard for URL changes - prevents out-of-order async resolution
     const urlChangeSequenceRef = useRef(0)
+    // The chat on screen, read by saves that run after a render or a wait
+    const currentSessionRef = useRef<ChatSession | null>(null)
+    // Goes up each time another chat is put on screen (creating the
+    // session of the chat on screen does not count)
+    const chatGenerationRef = useRef(0)
+    // Saves run one at a time, so two saves of a new chat create it once
+    const saveQueueRef = useRef<Promise<unknown>>(Promise.resolve())
+    // The last ticket number, and that of the newest data saved
+    const saveSeqRef = useRef(0)
+    const savedSeqRef = useRef(0)
+
+    const changeChat = useCallback((session: ChatSession | null) => {
+        chatGenerationRef.current++
+        currentSessionRef.current = session
+        setCurrentSession(session)
+        setCurrentSessionId(session?.id ?? null)
+    }, [])
 
     // Load sessions list
     const refreshSessions = useCallback(async () => {
@@ -101,14 +144,36 @@ export function useSessionManager(
                 // Load sessions list
                 const metadata = await getAllSessionMetadata()
                 setSessions(metadata)
+                // The desktop app may try its other port next launch, where
+                // an older version may have saved the chats: only when this
+                // origin surely has none (a failed read is not "none") and
+                // keeps no model settings or keys either
+                if (window.electronAPI?.chatsLoaded) {
+                    const count = await readSessionCount()
+                    // The app saves an empty config on its first load; the
+                    // providers are what holds the keys, besides an access
+                    // code
+                    let hasSettings = true
+                    try {
+                        const config = JSON.parse(
+                            localStorage.getItem(STORAGE_KEYS.modelConfigs) ??
+                                "{}",
+                        )
+                        hasSettings =
+                            (config.providers?.length ?? 0) > 0 ||
+                            !!localStorage.getItem(STORAGE_KEYS.accessCode)
+                    } catch {
+                        // Unreadable: treat as settings, and stay
+                    }
+                    if (count !== null && !hasSettings) {
+                        window.electronAPI.chatsLoaded(count).catch(() => {})
+                    }
+                }
 
                 // Only load a session if initialSessionId is provided (from URL param)
                 if (initialSessionId) {
                     const session = await getSession(initialSessionId)
-                    if (session) {
-                        setCurrentSession(session)
-                        setCurrentSessionId(session.id)
-                    }
+                    if (session) changeChat(session)
                     // If session not found, stay in blank state (URL has invalid session ID)
                 }
                 // If no initialSessionId, start with blank state (no auto-restore)
@@ -120,7 +185,7 @@ export function useSessionManager(
         }
 
         init()
-    }, [initialSessionId])
+    }, [initialSessionId, changeChat])
 
     // Handle URL session ID changes after initialization
     // Note: intentionally NOT including currentSessionId in deps to avoid race conditions
@@ -135,6 +200,7 @@ export function useSessionManager(
 
         async function handleSessionIdChange() {
             if (initialSessionId) {
+                const generation = chatGenerationRef.current
                 // URL has session ID - load it
                 const session = await getSession(initialSessionId)
 
@@ -143,16 +209,13 @@ export function useSessionManager(
                 if (currentSequence !== urlChangeSequenceRef.current) {
                     return
                 }
+                // Another chat was put on screen meanwhile (New Chat right
+                // after this one got its session id in the URL): keep it
+                if (generation !== chatGenerationRef.current) return
 
-                if (session) {
-                    // Only update if the session is different from current
-                    setCurrentSessionId((current) => {
-                        if (current !== session.id) {
-                            setCurrentSession(session)
-                            return session.id
-                        }
-                        return current
-                    })
+                // Only update if the session is different from current
+                if (session && currentSessionRef.current?.id !== session.id) {
+                    changeChat(session)
                 }
             }
             // Removed: else clause that clears session
@@ -161,11 +224,17 @@ export function useSessionManager(
         }
 
         handleSessionIdChange()
-    }, [initialSessionId, isAvailable])
+    }, [initialSessionId, isAvailable, changeChat])
 
-    // Refresh sessions on window focus (multi-tab sync)
+    // Refresh sessions on window focus (multi-tab sync), at most once per interval
+    const lastFocusRefreshRef = useRef(0)
     useEffect(() => {
         const handleFocus = () => {
+            const now = Date.now()
+            if (now - lastFocusRefreshRef.current < FOCUS_REFRESH_INTERVAL_MS) {
+                return
+            }
+            lastFocusRefreshRef.current = now
             refreshSessions()
         }
         window.addEventListener("focus", handleFocus)
@@ -177,9 +246,11 @@ export function useSessionManager(
         async (id: string): Promise<SessionData | null> => {
             if (id === currentSessionId) return null
 
-            // Save current session first if it has messages
-            if (currentSession && currentSession.messages.length > 0) {
-                await saveSession(currentSession)
+            // Save current session first if it has messages (as saved
+            // last: the caller may have just saved it)
+            const current = currentSessionRef.current
+            if (current && current.messages.length > 0) {
+                await saveSession(current)
             }
 
             // Load the target session
@@ -189,9 +260,7 @@ export function useSessionManager(
                 return null
             }
 
-            // Update state
-            setCurrentSession(session)
-            setCurrentSessionId(session.id)
+            changeChat(session)
 
             return {
                 messages: session.messages,
@@ -201,7 +270,7 @@ export function useSessionManager(
                 diagramHistory: session.diagramHistory,
             }
         },
-        [currentSessionId, currentSession],
+        [currentSessionId, changeChat],
     )
 
     // Delete a session
@@ -211,101 +280,137 @@ export function useSessionManager(
             await deleteSessionFromDB(id)
 
             // If deleting current session, clear state (caller will show new empty session)
-            if (wasCurrentSession) {
-                setCurrentSession(null)
-                setCurrentSessionId(null)
-            }
+            if (wasCurrentSession) changeChat(null)
 
             await refreshSessions()
 
             return { wasCurrentSession }
         },
-        [currentSessionId, refreshSessions],
+        [currentSessionId, refreshSessions, changeChat],
     )
 
     // Save current session data (debounced externally by caller)
-    // forSessionId: if provided, verify save targets correct session (prevents stale debounce writes)
     const saveCurrentSession = useCallback(
-        async (
-            data: SessionData,
-            forSessionId?: string | null,
-        ): Promise<void> => {
-            // If forSessionId is provided, verify it matches current session
-            // This prevents stale debounced saves from overwriting a newly switched session
-            if (
-                forSessionId !== undefined &&
-                forSessionId !== currentSessionId
-            ) {
-                return
+        (data: SessionData, ticket?: SaveTicket): Promise<boolean> => {
+            // The data is of the chat on screen when it was read
+            const { generation, seq } = ticket ?? {
+                generation: chatGenerationRef.current,
+                seq: ++saveSeqRef.current,
             }
+            const run = async (): Promise<boolean> => {
+                // That chat is no longer on screen (leaving it saved it)
+                if (generation !== chatGenerationRef.current) return true
+                // A copy read later was saved already (one that waited for
+                // its thumbnail must not undo it)
+                if (seq < savedSeqRef.current) return true
+                // Nothing can be stored without IndexedDB
+                if (!isIndexedDBAvailable()) return true
+                // The user may put another chat on screen while this one is
+                // written; the stored copy is still right, the state is not
+                const stillOnScreen = () =>
+                    chatGenerationRef.current === generation
+                const currentSession = currentSessionRef.current
 
-            if (!currentSession) {
-                // Create a new session if none exists
-                const newSession: ChatSession = {
-                    ...createEmptySession(),
+                if (!currentSession) {
+                    // Create a new session if none exists
+                    const newSession: ChatSession = {
+                        ...createEmptySession(),
+                        messages: data.messages,
+                        xmlSnapshots: data.xmlSnapshots,
+                        diagramXml: data.diagramXml,
+                        thumbnailDataUrl: data.thumbnailDataUrl,
+                        diagramHistory: data.diagramHistory,
+                        title: extractTitle(data.messages),
+                    }
+                    // Without a stored session, keep no session id (it would end
+                    // up in the URL and point to nothing after a reload)
+                    if (!(await saveSession(newSession))) {
+                        notifySaveFailed(dict.errors.sessionSaveFailed)
+                        return false
+                    }
+                    savedSeqRef.current = seq
+                    await enforceSessionLimit()
+                    if (stillOnScreen()) {
+                        currentSessionRef.current = newSession
+                        setCurrentSession(newSession)
+                        setCurrentSessionId(newSession.id)
+                    }
+                    await refreshSessions()
+                    return true
+                }
+
+                // Update existing session
+                const updatedSession: ChatSession = {
+                    ...currentSession,
                     messages: data.messages,
                     xmlSnapshots: data.xmlSnapshots,
                     diagramXml: data.diagramXml,
-                    thumbnailDataUrl: data.thumbnailDataUrl,
-                    diagramHistory: data.diagramHistory,
-                    title: extractTitle(data.messages),
+                    thumbnailDataUrl:
+                        data.thumbnailDataUrl ??
+                        currentSession.thumbnailDataUrl,
+                    diagramHistory:
+                        data.diagramHistory ?? currentSession.diagramHistory,
+                    updatedAt: Date.now(),
+                    // Update title if it's still default and we have messages
+                    title:
+                        currentSession.title === "New Chat" &&
+                        data.messages.length > 0
+                            ? extractTitle(data.messages)
+                            : currentSession.title,
                 }
-                await saveSession(newSession)
-                await enforceSessionLimit()
-                setCurrentSession(newSession)
-                setCurrentSessionId(newSession.id)
-                await refreshSessions()
-                return
+
+                if (!(await saveSession(updatedSession))) {
+                    notifySaveFailed(dict.errors.sessionSaveFailed)
+                    return false
+                }
+                savedSeqRef.current = seq
+                if (stillOnScreen()) {
+                    currentSessionRef.current = updatedSession
+                    setCurrentSession(updatedSession)
+                }
+
+                // Update sessions list metadata
+                setSessions((prev) =>
+                    prev.map((s) =>
+                        s.id === updatedSession.id
+                            ? {
+                                  ...s,
+                                  title: updatedSession.title,
+                                  updatedAt: updatedSession.updatedAt,
+                                  messageCount: updatedSession.messages.length,
+                                  hasDiagram:
+                                      !!updatedSession.diagramXml &&
+                                      updatedSession.diagramXml.trim().length >
+                                          0,
+                                  thumbnailDataUrl:
+                                      updatedSession.thumbnailDataUrl,
+                              }
+                            : s,
+                    ),
+                )
+                return true
             }
-
-            // Update existing session
-            const updatedSession: ChatSession = {
-                ...currentSession,
-                messages: data.messages,
-                xmlSnapshots: data.xmlSnapshots,
-                diagramXml: data.diagramXml,
-                thumbnailDataUrl:
-                    data.thumbnailDataUrl ?? currentSession.thumbnailDataUrl,
-                diagramHistory:
-                    data.diagramHistory ?? currentSession.diagramHistory,
-                updatedAt: Date.now(),
-                // Update title if it's still default and we have messages
-                title:
-                    currentSession.title === "New Chat" &&
-                    data.messages.length > 0
-                        ? extractTitle(data.messages)
-                        : currentSession.title,
-            }
-
-            await saveSession(updatedSession)
-            setCurrentSession(updatedSession)
-
-            // Update sessions list metadata
-            setSessions((prev) =>
-                prev.map((s) =>
-                    s.id === updatedSession.id
-                        ? {
-                              ...s,
-                              title: updatedSession.title,
-                              updatedAt: updatedSession.updatedAt,
-                              messageCount: updatedSession.messages.length,
-                              hasDiagram:
-                                  !!updatedSession.diagramXml &&
-                                  updatedSession.diagramXml.trim().length > 0,
-                              thumbnailDataUrl: updatedSession.thumbnailDataUrl,
-                          }
-                        : s,
-                ),
-            )
+            const result = saveQueueRef.current.then(run)
+            saveQueueRef.current = result.catch(() => {})
+            return result
         },
-        [currentSession, currentSessionId, refreshSessions],
+        [refreshSessions, dict],
     )
 
     // Clear current session state (for starting fresh without loading another session)
     const clearCurrentSession = useCallback(() => {
-        setCurrentSession(null)
-        setCurrentSessionId(null)
-    }, [])
+        changeChat(null)
+    }, [changeChat])
+
+    const getChatGeneration = useCallback(() => chatGenerationRef.current, [])
+
+    const getSaveTicket = useCallback(
+        (): SaveTicket => ({
+            generation: chatGenerationRef.current,
+            seq: ++saveSeqRef.current,
+        }),
+        [],
+    )
 
     return {
         sessions,
@@ -318,5 +423,7 @@ export function useSessionManager(
         saveCurrentSession,
         refreshSessions,
         clearCurrentSession,
+        getChatGeneration,
+        getSaveTicket,
     }
 }

hooks/use-validate-diagram.ts

@@ -6,6 +6,7 @@
 
 import { experimental_useObject as useObject } from "@ai-sdk/react"
 import { useCallback, useRef } from "react"
+import { getSelectedAIConfig } from "@/hooks/use-model-config"
 import { getApiEndpoint } from "@/lib/base-path"
 import {
     type ValidationResult,
@@ -39,6 +40,8 @@ export function useValidateDiagram(options: UseValidateDiagramOptions = {}) {
     const { object, submit, isLoading, error, stop } = useObject({
         api: getApiEndpoint("/api/validate-diagram"),
         schema: ValidationResultSchema,
+        // Resolved per request so a changed access code is picked up
+        headers: () => ({ "x-access-code": getSelectedAIConfig().accessCode }),
         onFinish: ({
             object,
             error: finishError,
@@ -100,9 +103,22 @@ export function useValidateDiagram(options: UseValidateDiagramOptions = {}) {
         [submit],
     )
 
+    /**
+     * End a running check (the user pressed Stop): its promise rejects with
+     * an AbortError, so the tool handler can finish at once.
+     */
+    const cancel = useCallback(() => {
+        const pending = pendingValidationRef.current
+        if (!pending) return
+        pendingValidationRef.current = null
+        stop()
+        pending.reject(new DOMException("Validation cancelled", "AbortError"))
+    }, [stop])
+
     /**
      * Validate with fallback - returns default valid result on error.
      * Use this to avoid blocking the user on validation failures.
+     * A cancelled check is passed on as its AbortError.
      */
     const validateWithFallback = useCallback(
         async (
@@ -112,6 +128,7 @@ export function useValidateDiagram(options: UseValidateDiagramOptions = {}) {
             try {
                 return await validate(imageData, sessionId)
             } catch (error) {
+                if ((error as Error)?.name === "AbortError") throw error
                 console.warn(
                     "[useValidateDiagram] Validation failed, using fallback:",
                     error,
@@ -127,6 +144,7 @@ export function useValidateDiagram(options: UseValidateDiagramOptions = {}) {
         validate,
         validateWithFallback,
         stop,
+        cancel,
 
         // State
         isValidating: isLoading,

lib/access-code.ts

@@ -0,0 +1,50 @@
+/**
+ * Refuse a POST that a page on another website could have sent. A browser
+ * sends a cross-site POST without asking first (CORS preflight) only with a
+ * text or form body, so the routes take JSON only. In the desktop app also
+ * refuse a foreign Host: a site that points its own domain name at
+ * 127.0.0.1 (DNS rebinding) is same-origin with the local server, but its
+ * requests carry that domain. A request the server builds itself has no
+ * Host. Returns the response to send, or null when the request may go on.
+ */
+export function rejectCrossSite(req: Request): Response | null {
+    const contentType = req.headers.get("content-type") ?? ""
+    if (!/^\s*application\/json\b/i.test(contentType)) {
+        return Response.json(
+            { error: "Content-Type must be application/json" },
+            { status: 415 },
+        )
+    }
+    const host = req.headers.get("host")
+    if (
+        process.env.NEXT_AI_DRAWIO_DESKTOP === "1" &&
+        host &&
+        !/^(127\.0\.0\.1|localhost)(:\d+)?$/i.test(host)
+    ) {
+        return Response.json({ error: "Forbidden" }, { status: 403 })
+    }
+    return null
+}
+
+/**
+ * Check the x-access-code header against ACCESS_CODE_LIST.
+ * Returns a 401 response to send back when the check fails, or null when the
+ * request may continue (including when no access codes are configured).
+ */
+export function checkAccessCode(req: Request): Response | null {
+    const accessCodes =
+        process.env.ACCESS_CODE_LIST?.split(",")
+            .map((code) => code.trim())
+            .filter(Boolean) || []
+    if (accessCodes.length === 0) return null
+
+    const accessCodeHeader = req.headers.get("x-access-code")
+    if (accessCodeHeader && accessCodes.includes(accessCodeHeader)) return null
+
+    return Response.json(
+        {
+            error: "Invalid or missing access code. Please configure it in Settings.",
+        },
+        { status: 401 },
+    )
+}

lib/admin/providers.ts

@@ -2,6 +2,7 @@ import { z } from "zod"
 import {
     ProviderNameSchema,
     type ServerModelsConfig,
+    slugify,
 } from "@/lib/server-model-config"
 import {
     FIXED_CRED_PROVIDERS,
@@ -182,12 +183,15 @@ export function validateAdminProviders(
             return `${PROVIDER_INFO[single].label} is already configured in AI_MODELS_CONFIG / ai-models.json and shares global credentials. Manage it via the environment configuration instead.`
         }
     }
+    // Server model ids are built from the slugified name, so names must
+    // stay distinct after slugifying ("OpenAI" and "openai" would collide)
     const names = list.map((p) => displayName(p))
-    if (new Set(names).size !== names.length) {
-        return "Provider display names must be unique."
+    const slugs = names.map(slugify)
+    if (new Set(slugs).size !== slugs.length) {
+        return "Provider display names must be unique (ignoring case and punctuation)."
     }
-    const envNames = new Set(envProviders.map((p) => p.name))
-    const clash = names.find((n) => envNames.has(n))
+    const envSlugs = new Set(envProviders.map((p) => slugify(p.name)))
+    const clash = names.find((_, i) => envSlugs.has(slugs[i]))
     if (clash) {
         return `"${clash}" is already defined in AI_MODELS_CONFIG / ai-models.json. Use a different display name.`
     }
@@ -240,10 +244,14 @@ export function deriveEnvUpdates(
         indexByProvider.set(p.provider, index + 1)
 
         if (p.provider === "bedrock") {
-            if (p.awsAccessKeyId) updates.AWS_ACCESS_KEY_ID = p.awsAccessKeyId
+            // ADMIN_ names keep the standard AWS_* vars untouched, so other
+            // AWS clients (e.g. the DynamoDB quota table) keep their own
+            // credentials instead of picking up the panel's Bedrock keys
+            if (p.awsAccessKeyId)
+                updates.ADMIN_AWS_ACCESS_KEY_ID = p.awsAccessKeyId
             if (p.awsSecretAccessKey)
-                updates.AWS_SECRET_ACCESS_KEY = p.awsSecretAccessKey
-            if (p.awsRegion) updates.AWS_REGION = p.awsRegion
+                updates.ADMIN_AWS_SECRET_ACCESS_KEY = p.awsSecretAccessKey
+            if (p.awsRegion) updates.ADMIN_AWS_REGION = p.awsRegion
         } else if (p.provider === "vertexai") {
             if (p.vertexApiKey) updates.GOOGLE_VERTEX_API_KEY = p.vertexApiKey
             if (p.baseUrl) updates.GOOGLE_VERTEX_BASE_URL = p.baseUrl
@@ -284,6 +292,10 @@ function derivedEnvKeys(list: StoredAdminProvider[]): string[] {
         const index = indexByProvider.get(p.provider) ?? 0
         indexByProvider.set(p.provider, index + 1)
         if (p.provider === "bedrock") {
+            keys.add("ADMIN_AWS_ACCESS_KEY_ID")
+            keys.add("ADMIN_AWS_SECRET_ACCESS_KEY")
+            keys.add("ADMIN_AWS_REGION")
+            // Written by older versions; listed so the next save clears them
             keys.add("AWS_ACCESS_KEY_ID")
             keys.add("AWS_SECRET_ACCESS_KEY")
             keys.add("AWS_REGION")

lib/admin/settings.ts

@@ -10,13 +10,27 @@ interface SettingsFile {
     values: Record<string, string>
 }
 
-// Original env values snapshotted before the first overlay, so removing a
-// key from the settings file restores the env default. null = was unset.
-const originalEnv: Record<string, string | null> = {}
-// Keys currently overlaid, so we can restore ones removed from the file.
-let overlaidKeys = new Set<string>()
+interface SettingsState {
+    // Original env values snapshotted before the first overlay, so removing
+    // a key from the settings file restores the env default. null = was unset.
+    originalEnv: Record<string, string | null>
+    // Keys currently overlaid, so we can restore ones removed from the file.
+    overlaidKeys: Set<string>
+    cachedSettings: Record<string, string> | null
+}
 
-let cachedSettings: Record<string, string> | null = null
+// Kept on globalThis because the build can load this module more than once
+// (instrumentation.ts and the API routes get separate copies); per-module
+// state would make a route forget what instrumentation overlaid at startup.
+const globalState = globalThis as typeof globalThis & {
+    __adminSettingsState?: SettingsState
+}
+globalState.__adminSettingsState ??= {
+    originalEnv: {},
+    overlaidKeys: new Set(),
+    cachedSettings: null,
+}
+const state = globalState.__adminSettingsState
 
 export function getSettingsPath(): string {
     const custom = process.env.SETTINGS_FILE
@@ -25,7 +39,7 @@ export function getSettingsPath(): string {
 }
 
 export function loadSettings(): Record<string, string> {
-    if (cachedSettings) return cachedSettings
+    if (state.cachedSettings) return state.cachedSettings
     try {
         const raw = fs.readFileSync(getSettingsPath(), "utf8")
         const parsed = JSON.parse(raw) as SettingsFile
@@ -43,21 +57,22 @@ export function loadSettings(): Record<string, string> {
         for (const [key, value] of Object.entries(rawValues)) {
             if (typeof value === "string") values[key] = value
         }
-        cachedSettings = values
+        state.cachedSettings = values
     } catch (err: any) {
         if (err?.code !== "ENOENT") {
             console.error("[admin-settings] Failed to read settings file:", err)
         }
-        cachedSettings = {}
+        state.cachedSettings = {}
     }
-    return cachedSettings
+    return state.cachedSettings
 }
 
 export function applyToEnv(): void {
     const values = loadSettings()
+    const { originalEnv } = state
 
     // Restore env for keys that were overlaid before but are now gone
-    for (const key of overlaidKeys) {
+    for (const key of state.overlaidKeys) {
         if (!(key in values)) {
             const original = originalEnv[key]
             if (original === null) delete process.env[key]
@@ -72,12 +87,12 @@ export function applyToEnv(): void {
         process.env[key] = value
     }
 
-    overlaidKeys = new Set(Object.keys(values))
+    state.overlaidKeys = new Set(Object.keys(values))
 }
 
 // The effective env value if the file entry were removed (for fallback display)
 export function getEnvFallback(key: string): string | null {
-    if (overlaidKeys.has(key)) return originalEnv[key] ?? null
+    if (state.overlaidKeys.has(key)) return state.originalEnv[key] ?? null
     return process.env[key] ?? null
 }
 
@@ -101,7 +116,7 @@ export function saveSettings(updates: Record<string, string | null>): void {
     fs.writeFileSync(tmpPath, JSON.stringify(data, null, 2), { mode: 0o600 })
     fs.renameSync(tmpPath, filePath)
 
-    cachedSettings = current
+    state.cachedSettings = current
     applyToEnv()
 }
 
@@ -110,7 +125,12 @@ let writableCache: boolean | null = null
 export function isSettingsWritable(): boolean {
     if (writableCache !== null) return writableCache
     try {
-        const dir = path.dirname(getSettingsPath())
+        const filePath = getSettingsPath()
+        if (fs.existsSync(filePath) && !fs.statSync(filePath).isFile()) {
+            writableCache = false
+            return writableCache
+        }
+        const dir = path.dirname(filePath)
         fs.mkdirSync(dir, { recursive: true })
         fs.accessSync(dir, fs.constants.W_OK)
         writableCache = true
@@ -122,13 +142,13 @@ export function isSettingsWritable(): boolean {
 
 // Test-only: reset module state
 export function _resetForTests(): void {
-    cachedSettings = null
+    state.cachedSettings = null
     writableCache = null
-    for (const key of overlaidKeys) {
-        const original = originalEnv[key]
+    for (const key of state.overlaidKeys) {
+        const original = state.originalEnv[key]
         if (original === null) delete process.env[key]
         else if (original !== undefined) process.env[key] = original
     }
-    overlaidKeys = new Set()
-    for (const key of Object.keys(originalEnv)) delete originalEnv[key]
+    state.overlaidKeys = new Set()
+    state.originalEnv = {}
 }

lib/aihubmix-models.ts

@@ -1,79 +0,0 @@
-export const AIHUBMIX_MODELS_ENDPOINT = "https://aihubmix.com/api/v1/models"
-
-const NON_CHAT_MODEL_TYPES = new Set([
-    "embedding",
-    "image_generation",
-    "rerank",
-    "transcription",
-    "tts",
-    "video",
-])
-
-type AihubmixModelListPayload = {
-    data?: unknown
-}
-
-type AihubmixModelRecord = {
-    model_id?: unknown
-    types?: unknown
-}
-
-function getModelTypes(types: unknown): Set<string> {
-    if (typeof types !== "string") {
-        return new Set()
-    }
-
-    return new Set(
-        types
-            .split(",")
-            .map((type) => type.trim())
-            .filter(Boolean),
-    )
-}
-
-function isChatModel(record: AihubmixModelRecord): record is {
-    model_id: string
-    types: string
-} {
-    if (typeof record.model_id !== "string" || !record.model_id.trim()) {
-        return false
-    }
-
-    const types = getModelTypes(record.types)
-    if (!types.has("llm")) {
-        return false
-    }
-
-    return !Array.from(NON_CHAT_MODEL_TYPES).some((type) => types.has(type))
-}
-
-export function extractAihubmixModelIds(payload: unknown): string[] {
-    const data = (payload as AihubmixModelListPayload)?.data
-    if (!Array.isArray(data)) {
-        return []
-    }
-
-    const seen = new Set<string>()
-    const modelIds: string[] = []
-
-    for (const item of data) {
-        if (!item || typeof item !== "object") {
-            continue
-        }
-
-        const record = item as AihubmixModelRecord
-        if (!isChatModel(record)) {
-            continue
-        }
-
-        const modelId = record.model_id.trim()
-        if (seen.has(modelId)) {
-            continue
-        }
-
-        seen.add(modelId)
-        modelIds.push(modelId)
-    }
-
-    return modelIds
-}

lib/cached-responses.ts

@@ -1,6 +1,8 @@
 export interface CachedResponse {
     promptText: string
     hasImage: boolean
+    // Name of the bundled example file the prompt is sent with
+    fileName?: string
     xml: string
 }
 
@@ -254,6 +256,7 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
     {
         promptText: "Replicate this in aws style",
         hasImage: true,
+        fileName: "architecture.png",
         xml: `<mxCell id="2" value="AWS" style="sketch=0;outlineConnect=0;gradientColor=none;html=1;whiteSpace=wrap;fontSize=12;fontStyle=0;container=1;pointerEvents=0;collapsible=0;recursiveResize=0;shape=mxgraph.aws4.group;grIcon=mxgraph.aws4.group_aws_cloud;strokeColor=#232F3E;fillColor=none;verticalAlign=top;align=left;spacingLeft=30;fontColor=#232F3E;dashed=0;rounded=1;arcSize=5;" vertex="1" parent="1">
     <mxGeometry x="340" y="40" width="880" height="520" as="geometry"/>
   </mxCell>
@@ -318,6 +321,7 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
     {
         promptText: "Replicate this flowchart.",
         hasImage: true,
+        fileName: "example.png",
         xml: `<mxCell id="2" value="Lamp doesn't work" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#ffcccc;strokeColor=#000000;strokeWidth=2;fontSize=18;fontStyle=0;" vertex="1" parent="1">
     <mxGeometry x="140" y="40" width="180" height="60" as="geometry"/>
   </mxCell>
@@ -379,6 +383,7 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
     {
         promptText: "Summarize this paper as a diagram",
         hasImage: true,
+        fileName: "chain-of-thought.txt",
         xml: `<mxCell id="title_bg" parent="1"
                     style="rounded=1;whiteSpace=wrap;html=1;fillColor=#1a237e;strokeColor=none;arcSize=8;"
                     value="" vertex="1">
@@ -879,14 +884,19 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
     },
 ]
 
+// Examples that come with a file only match when that exact example file is
+// attached, so a user's own file with the same prompt still goes to the model.
+// Callers that can't tell file names (the server) only get text-only examples.
 export function findCachedResponse(
     promptText: string,
     hasImage: boolean,
+    fileName?: string,
 ): CachedResponse | undefined {
     return CACHED_EXAMPLE_RESPONSES.find(
         (c) =>
             c.promptText === promptText &&
             c.hasImage === hasImage &&
+            (!c.fileName || c.fileName === fileName) &&
             c.xml !== "",
     )
 }

lib/chat-helpers.ts

@@ -6,25 +6,37 @@ export const MAX_FILE_SIZE = 2 * 1024 * 1024 // 2MB
 export const MAX_FILES = 5
 
 // Helper function to validate file parts in messages
+// Checks every message, since history is sent to the model too
 export function validateFileParts(messages: any[]): {
     valid: boolean
     error?: string
 } {
-    const lastMessage = messages[messages.length - 1]
-    const fileParts =
-        lastMessage?.parts?.filter((p: any) => p.type === "file") || []
+    for (const message of messages) {
+        const fileParts =
+            message?.parts?.filter((p: any) => p.type === "file") || []
 
-    if (fileParts.length > MAX_FILES) {
-        return {
-            valid: false,
-            error: `Too many files. Maximum ${MAX_FILES} allowed.`,
+        if (fileParts.length > MAX_FILES) {
+            return {
+                valid: false,
+                error: `Too many files. Maximum ${MAX_FILES} allowed.`,
+            }
         }
-    }
 
-    for (const filePart of fileParts) {
-        // Data URLs format: data:image/png;base64,<data>
-        // Base64 increases size by ~33%, so we check the decoded size
-        if (filePart.url?.startsWith("data:")) {
+        for (const filePart of fileParts) {
+            // The client sends files inline. Any other URL would be downloaded
+            // by the server (AI SDK does that for models without URL support).
+            if (
+                typeof filePart.url !== "string" ||
+                !filePart.url.startsWith("data:")
+            ) {
+                return {
+                    valid: false,
+                    error: "Files must be uploaded inline as data URLs.",
+                }
+            }
+
+            // Data URLs format: data:image/png;base64,<data>
+            // Base64 increases size by ~33%, so we check the decoded size
             const base64Data = filePart.url.split(",")[1]
             if (base64Data) {
                 const sizeInBytes = Math.ceil((base64Data.length * 3) / 4)
@@ -41,49 +53,83 @@ export function validateFileParts(messages: any[]): {
     return { valid: true }
 }
 
-// Helper function to check if diagram is minimal/empty
-export function isMinimalDiagram(xml: string): boolean {
-    const stripped = xml.replace(/\s/g, "")
-    return !stripped.includes('id="2"')
+// A tool-call input providers accept: a non-empty JSON object
+function isValidToolInput(input: unknown): boolean {
+    return !!input && typeof input === "object" && Object.keys(input).length > 0
 }
 
 // Helper function to replace historical tool call XML with placeholders
 // This reduces token usage and forces LLM to rely on the current diagram XML (source of truth)
-// Also fixes invalid/undefined inputs from interrupted streaming
+// Tool calls with invalid inputs are left for dropInvalidToolCalls to remove
 export function replaceHistoricalToolInputs(messages: any[]): any[] {
     return messages.map((msg) => {
         if (msg.role !== "assistant" || !Array.isArray(msg.content)) {
             return msg
         }
-        const replacedContent = msg.content
-            .map((part: any) => {
-                if (part.type === "tool-call") {
-                    const toolName = part.toolName
-                    // Fix invalid/undefined inputs from interrupted streaming
-                    if (
-                        !part.input ||
-                        typeof part.input !== "object" ||
-                        Object.keys(part.input).length === 0
-                    ) {
-                        // Skip tool calls with invalid inputs entirely
-                        return null
-                    }
-                    if (
-                        toolName === "display_diagram" ||
-                        toolName === "edit_diagram"
-                    ) {
-                        return {
-                            ...part,
-                            input: {
-                                placeholder:
-                                    "[XML content replaced - see current diagram XML in system context]",
-                            },
-                        }
-                    }
+        const replacedContent = msg.content.map((part: any) => {
+            if (
+                part.type === "tool-call" &&
+                isValidToolInput(part.input) &&
+                (part.toolName === "display_diagram" ||
+                    part.toolName === "edit_diagram")
+            ) {
+                return {
+                    ...part,
+                    input: {
+                        placeholder:
+                            "[XML content replaced - see current diagram XML in system context]",
+                    },
                 }
-                return part
-            })
-            .filter(Boolean) // Remove null entries (invalid tool calls)
+            }
+            return part
+        })
         return { ...msg, content: replacedContent }
     })
 }
+
+// Remove tool-calls with invalid inputs (from failed repair or interrupted streaming),
+// together with their tool-results: providers reject a result whose call is missing.
+// Messages left empty are removed too (Bedrock rejects empty content arrays).
+export function dropInvalidToolCalls(messages: any[]): any[] {
+    const droppedIds = new Set<string>()
+    return messages
+        .map((msg) => {
+            if (!Array.isArray(msg.content)) return msg
+            const content = msg.content.filter((part: any) => {
+                if (
+                    msg.role === "assistant" &&
+                    part.type === "tool-call" &&
+                    !isValidToolInput(part.input)
+                ) {
+                    console.warn(
+                        `[chat-helpers] Dropping tool-call with invalid input:`,
+                        { toolName: part.toolName, input: part.input },
+                    )
+                    droppedIds.add(part.toolCallId)
+                    return false
+                }
+                // Results always come after their call, so the id is known by now
+                return !(
+                    part.type === "tool-result" &&
+                    droppedIds.has(part.toolCallId)
+                )
+            })
+            return { ...msg, content }
+        })
+        .filter((msg) => !Array.isArray(msg.content) || msg.content.length > 0)
+}
+
+// Fix common LLM JSON mistakes in tool-call input before jsonrepair runs
+export function fixToolInputJson(input: string): string {
+    return (
+        input
+            // Inconsistent quote escaping in XML attributes inside JSON strings:
+            // y="-20\" (opening quote unescaped, closing escaped) becomes y=\"-20\".
+            // Must run before the key fix below, which would rewrite the `="`.
+            .replace(/(\w+)="([^"]*?)\\"/g, '$1=\\"$2\\"')
+            // `:=` instead of `: `
+            .replace(/:=/g, ": ")
+            // `"key"= "` instead of `"key": "`, only for JSON keys
+            .replace(/"(\w+)"\s*=\s*"/g, '"$1": "')
+    )
+}

lib/deprecated-params.ts

@@ -0,0 +1,94 @@
+import { wrapLanguageModel } from "ai"
+import { rejectionText } from "@/lib/output-token-limit"
+
+type WrappedModel = ReturnType<typeof wrapLanguageModel>
+
+/**
+ * Claude 4.7 and later answer a non-default temperature, top_p or top_k,
+ * and the extended thinking budget (thinking type "enabled"), with a 400.
+ * TEMPERATURE and the *_THINKING_BUDGET_TOKENS settings send exactly these.
+ */
+const DEPRECATED_PARAM =
+    /`?(?:temperature|top_p|top_k)`? is deprecated for this model|"?thinking\.type\.enabled"? is not supported/i
+
+interface CallParams {
+    temperature?: number
+    topP?: number
+    topK?: number
+    providerOptions?: Record<string, Record<string, unknown> | undefined>
+}
+
+// What these models take instead of a budget. Without display "summarized"
+// they think but send no thinking text to show.
+const ADAPTIVE_THINKING = { type: "adaptive", display: "summarized" }
+
+/** Turn a thinking config of type "enabled" stored under key into adaptive */
+function adaptiveThinking(
+    options: Record<string, unknown> | undefined,
+    key: string,
+): Record<string, unknown> | undefined {
+    const config = options?.[key] as { type?: string } | undefined
+    if (config?.type !== "enabled") return options
+    return { ...options, [key]: ADAPTIVE_THINKING }
+}
+
+/**
+ * The params without the settings newer Claude models reject, or null when
+ * the error is about something else or there is nothing to change. The
+ * model then runs with its default sampling, and a thinking budget becomes
+ * adaptive thinking.
+ */
+export function withoutDeprecatedParams<T extends CallParams>(
+    error: unknown,
+    params: T,
+): T | null {
+    const text = rejectionText(error)
+    if (!text || !DEPRECATED_PARAM.test(text)) return null
+
+    const { temperature, topP, topK, ...rest } = params
+    const options = params.providerOptions
+    const anthropic = adaptiveThinking(options?.anthropic, "thinking")
+    const bedrock = adaptiveThinking(options?.bedrock, "reasoningConfig")
+    const changed =
+        temperature !== undefined ||
+        topP !== undefined ||
+        topK !== undefined ||
+        anthropic !== options?.anthropic ||
+        bedrock !== options?.bedrock
+    if (!changed) return null
+
+    return {
+        ...rest,
+        ...(options && {
+            providerOptions: {
+                ...options,
+                ...(anthropic && { anthropic }),
+                ...(bedrock && { bedrock }),
+            },
+        }),
+    } as T
+}
+
+/** Retry the stream once without the settings newer Claude models reject. */
+export function withDeprecatedParamsFallback(
+    model: WrappedModel,
+): WrappedModel {
+    return wrapLanguageModel({
+        model,
+        middleware: {
+            specificationVersion: "v3",
+            async wrapStream({ doStream, params, model: inner }) {
+                try {
+                    return await doStream()
+                } catch (error) {
+                    const retry = withoutDeprecatedParams(error, params)
+                    if (!retry) throw error
+                    console.warn(
+                        "[model params] Rejected sampling or thinking settings, retrying with default sampling and adaptive thinking",
+                    )
+                    return await inner.doStream(retry)
+                }
+            },
+        },
+    })
+}

lib/diagram-engine/catalog.ts

@@ -1,219 +0,0 @@
-/**
- * The stencil catalog: a name → verbatim draw.io style map.
- *
- * This is the anti-hallucination layer. The model asks for `icon("s3")`; the engine
- * looks the name up here and gets the exact style draw.io ships, including the official
- * category colour, the connection points and `aspect=fixed`. A name that is not in the
- * catalog fails at build time with a suggestion, rather than becoming an empty square in
- * the rendered diagram — which is what happens when a model writes
- * `resIcon=mxgraph.aws4.s3_bucket_thing` by hand and nothing checks it.
- *
- * The styles are verbatim from draw.io's own shape index (via drawio-ai-kit, which
- * generated them from jgraph/drawio-mcp's index, Apache-2.0). Nothing here is
- * hand-assembled, so there is no chance of a plausible-looking but wrong colour.
- */
-
-import stencils from "./data/aws-stencils.json"
-
-const ICONS = stencils.icons as Record<string, string>
-const GROUPS = stencils.groups as Record<string, string>
-
-export interface CatalogEntry {
-    name: string
-    kind: "icon" | "group"
-    style: string
-    /** Official colour from the style, for showing the model what it is getting. */
-    color: string | null
-}
-
-function colorOf(style: string): string | null {
-    return style.match(/(?:^|;)fillColor=([^;]+)/)?.[1] ?? null
-}
-
-/** Exact lookup. Returns null for an unknown name — never a guess. */
-export function lookupStencil(
-    name: string,
-    kind?: "icon" | "group",
-): CatalogEntry | null {
-    if (kind !== "group" && ICONS[name])
-        return {
-            name,
-            kind: "icon",
-            style: ICONS[name],
-            color: colorOf(ICONS[name]),
-        }
-    if (kind !== "icon" && GROUPS[name])
-        return {
-            name,
-            kind: "group",
-            style: GROUPS[name],
-            color: colorOf(GROUPS[name]),
-        }
-    return null
-}
-
-/** The resolver the renderer takes, so the engine itself does not depend on the catalog. */
-export function resolveStyle(
-    name: string,
-    kind: "icon" | "group",
-): string | null {
-    return lookupStencil(name, kind)?.style ?? null
-}
-
-/** Normalise for matching: lowercase, and non-alphanumerics collapsed to single spaces. */
-function norm(s: string): string {
-    return s
-        .toLowerCase()
-        .replace(/[^a-z0-9]+/g, " ")
-        .trim()
-}
-
-/**
- * Shorthand people type, mapped to words that actually appear in a catalog name.
- *
- * The direction matters: the target has to exist in the catalog. AWS's own stencil names
- * are already abbreviated — EKS is `eks`, not `elastic_kubernetes_service`, and nothing
- * in the catalog contains the word "kubernetes" at all — so expanding an abbreviation
- * into its full product name finds nothing. These entries go the other way, from a
- * spoken-out name or a nickname to the token the catalog uses.
- */
-const ALIASES: Record<string, string> = {
-    k8s: "eks",
-    kubernetes: "eks",
-    kube: "eks",
-    alb: "application load balancer",
-    nlb: "network load balancer",
-    elb: "elastic load balancing",
-    asg: "auto scaling",
-    apigw: "api gateway",
-    cf: "cloudfront",
-    cw: "cloudwatch",
-    ddb: "dynamodb",
-    tgw: "transit gateway",
-    igw: "internet gateway",
-    r53: "route 53",
-    iam: "identity and access management",
-    kms: "key management service",
-    postgres: "rds",
-    postgresql: "rds",
-    mysql: "rds",
-    aurora: "aurora",
-    bucket: "s3",
-}
-
-/**
- * Score one entry against the query tokens. Higher is better; 0 means no match.
- *
- * The extra-words penalty is what makes "s3" return `s3` rather than
- * `backup_aws_backup_support_for_amazon_s3` — both contain the token, so without it the
- * winner comes down to iteration order. It counts only the words the query did NOT ask
- * for, so a deliberately multi-word query like "nat gateway" is not punished for being
- * specific.
- */
-function score(name: string, qTokens: string[], qJoined: string): number {
-    const n = norm(name)
-    const words = n.split(" ")
-    let s = 0
-    if (n === qJoined) s += 100
-    if (n.replace(/ /g, "") === qJoined.replace(/ /g, "")) s += 60
-    for (const t of qTokens) {
-        if (words.includes(t)) s += 25
-        else if (n.includes(t)) s += 12
-    }
-    if (s === 0) return 0
-    const extra = words.filter((w) => !qTokens.includes(w)).length
-    return s - Math.min(24, extra * 4)
-}
-
-export interface SearchHit {
-    name: string
-    kind: "icon" | "group"
-    color: string | null
-}
-
-/**
- * Find stencils by keyword.
- *
- * Returns names and colours only, not styles. The model builds with `icon("<name>")` and
- * the engine resolves the style itself, so sending the style — around 600 characters per
- * AWS entry, and 20KB+ for an Azure one with an embedded image — would be pure context
- * burn.
- */
-export function searchStencils(
-    query: string,
-    opts: { limit?: number; kind?: "icon" | "group" } = {},
-): SearchHit[] {
-    const limit = opts.limit ?? 8
-    const tokens = norm(query)
-        .split(" ")
-        .filter(Boolean)
-        .map((t) => ALIASES[t] ?? t)
-        .flatMap((t) => t.split(" "))
-    if (tokens.length === 0) return []
-    const joined = tokens.join(" ")
-
-    const pool: [string, string, "icon" | "group"][] = []
-    if (opts.kind !== "group")
-        for (const [n, st] of Object.entries(ICONS)) pool.push([n, st, "icon"])
-    if (opts.kind !== "icon")
-        for (const [n, st] of Object.entries(GROUPS))
-            pool.push([n, st, "group"])
-
-    return pool
-        .map(([name, style, kind]) => ({
-            name,
-            kind,
-            color: colorOf(style),
-            s: score(name, tokens, joined),
-        }))
-        .filter((r) => r.s > 0)
-        .sort((a, b) => b.s - a.s || a.name.length - b.name.length)
-        .slice(0, limit)
-        .map(({ name, kind, color }) => ({ name, kind, color }))
-}
-
-/**
- * Suggest real names for one that does not exist.
- *
- * Plain search is not quite the right tool here. A model that writes
- * `s3_bucket_storage` most likely meant `s3`, but searching that whole phrase ranks
- * `s3_storage_lens` first — it matches more of the query. So we also search the
- * leading token on its own and put those hits first: an invented name is usually a
- * real service name with extra words stuck on the end.
- */
-function suggestFor(name: string, kind: "icon" | "group"): string[] {
-    const words = norm(name.replace(/_/g, " ")).split(" ").filter(Boolean)
-    const out: string[] = []
-    const add = (hits: SearchHit[]) => {
-        for (const h of hits) if (!out.includes(h.name)) out.push(h.name)
-    }
-    if (words.length > 1) add(searchStencils(words[0], { limit: 2, kind }))
-    add(searchStencils(words.join(" "), { limit: 3, kind }))
-    return out.slice(0, 3)
-}
-
-/**
- * Validate the icon names in a tree before laying it out, so a bad name is reported as
- * a correctable error with suggestions instead of rendering as a blank square — which is
- * what an unchecked invented name becomes in draw.io.
- */
-export function checkNames(
-    names: { id: string; name: string; kind: "icon" | "group" }[],
-): { id: string; name: string; suggestions: string[] }[] {
-    const bad: { id: string; name: string; suggestions: string[] }[] = []
-    for (const n of names) {
-        if (!n.name || lookupStencil(n.name, n.kind)) continue
-        bad.push({
-            id: n.id,
-            name: n.name,
-            suggestions: suggestFor(n.name, n.kind),
-        })
-    }
-    return bad
-}
-
-/** Total catalog size, for the tool description. */
-export const CATALOG_SIZE = {
-    icons: Object.keys(ICONS).length,
-    groups: Object.keys(GROUPS).length,
-}

lib/diagram-engine/graph.ts

@@ -1,393 +0,0 @@
-/**
- * Graph → layers. What turns a flat list of nodes and arrows into a diagram.
- *
- * The engine's layout can only arrange what nesting tells it to: a container stacks its
- * children in one direction, so six boxes declared in a row become six boxes in a row. For
- * a flowchart that is the wrong answer, and measurably so — an order-approval flow declared
- * in its natural order comes out as one column, which forces the arrow from the decision to
- * its second branch to jump over the first branch, and the arrow to the merge point to jump
- * back over that. The layout never looked at the arrows.
- *
- * This computes what it should have looked at. Three steps, the standard shape of a layered
- * graph drawing (Sugiyama's algorithm):
- *
- *   1. LAYER — how far along the flow each node sits. Longest path from a source, so an
- *      arrow always points forwards and no arrow skips backwards through a layer.
- *   2. ORDER — who goes left and who goes right within a layer. Chosen to reduce the number
- *      of arrows that cross, which is what makes a flowchart readable.
- *   3. EMIT — one invisible row container per layer, which the existing layout then places.
- *
- * Step 3 is why this file is small: the coordinate work already exists, and it is the same
- * code that lays out an AWS diagram. What was missing was only the decision of what goes in
- * which row.
- */
-
-import type { Operation } from "./operations"
-import type { BoxShape, Role } from "./types"
-
-/** A node in the graph the caller wants drawn. */
-export interface GraphNode {
-    id: string
-    label: string
-    /** Flowchart outline. `decision` for a branch, `terminator` for a start or end point. */
-    shape?: BoxShape
-    /** Catalog stencil name. When set the node renders as an icon rather than a box. */
-    icon?: string
-    /**
-     * Semantic group name, e.g. "remote" or "local". Nodes sharing a group get the same
-     * fill colour from the engine's palette, assigned in order of first appearance — the
-     * caller names the grouping and never touches a colour.
-     */
-    group?: string
-    /** Information role (heading, callout, metric…); the theme decides how it looks. */
-    role?: Role
-}
-
-/** An arrow. Direction matters: it is what determines the layering. */
-export interface GraphEdge {
-    source: string
-    target: string
-    label?: string
-    dashed?: boolean
-    /** Thick coloured arrow for THE key relationship. */
-    bold?: boolean
-    /** Arrowhead tokens, passed through — see LinkSpec. */
-    head?: string
-    tail?: string
-    headFill?: boolean
-    tailFill?: boolean
-}
-
-export interface GraphOptions {
-    /** "col" (default): layers stack downwards. "row": layers run left to right. */
-    flow?: "col" | "row"
-    /** Container to embed the graph in; absent means the page. */
-    parent?: string
-    /**
-     * Namespace for the synthetic layer-container ids. Without one, two graphs on one
-     * page would both emit `__layers`/`__layer0` and the second would be rejected as a
-     * duplicate id.
-     */
-    prefix?: string
-    /** Id for the outer container itself; defaults to `${prefix}__layers`. */
-    rootId?: string
-}
-
-/** Distance between layers. */
-const LAYER_GAP = 48
-/** Distance between nodes within a layer. */
-const NODE_GAP = 60
-/** Prefix for the generated layer container ids. */
-const LAYER_ID = "__layer"
-
-export interface GraphResult {
-    operations: Operation[]
-    /** The nodes of each layer, in the order they were placed. */
-    layers: string[][]
-    /** Edges dropped because an endpoint is not in the node list. */
-    unknownEndpoints: string[]
-    /** Edges that had to be treated as loops rather than as layering constraints. */
-    backEdges: { source: string; target: string }[]
-}
-
-/**
- * Break every cycle, so the graph can be layered at all.
- *
- * A depth-first walk; any arrow pointing at a node still on the current path is a way back
- * to where we came from, and cannot be a "this comes after that" constraint. Those arrows
- * are still DRAWN — a review loop is the point of the diagram — they just do not get a say
- * in which layer anything lands in.
- */
-function breakCycles(
-    nodes: string[],
-    edges: GraphEdge[],
-): { forward: GraphEdge[]; back: GraphEdge[] } {
-    const out = new Map<string, GraphEdge[]>(nodes.map((n) => [n, []]))
-    for (const e of edges) out.get(e.source)?.push(e)
-
-    const forward: GraphEdge[] = []
-    const back: GraphEdge[] = []
-    const onPath = new Set<string>()
-    const done = new Set<string>()
-
-    // An explicit stack, not recursion: a 500-node dependency graph is a plausible input and
-    // a recursive walk over one would overflow.
-    for (const root of nodes) {
-        if (done.has(root)) continue
-        const stack: { id: string; next: number }[] = [{ id: root, next: 0 }]
-        onPath.add(root)
-        while (stack.length > 0) {
-            const top = stack[stack.length - 1]
-            const list = out.get(top.id) ?? []
-            if (top.next >= list.length) {
-                onPath.delete(top.id)
-                done.add(top.id)
-                stack.pop()
-                continue
-            }
-            const e = list[top.next++]
-            if (onPath.has(e.target)) {
-                back.push(e)
-                continue
-            }
-            forward.push(e)
-            if (!done.has(e.target)) {
-                onPath.add(e.target)
-                stack.push({ id: e.target, next: 0 })
-            }
-        }
-    }
-    return { forward, back }
-}
-
-/**
- * Assign each node to a layer: the longest path to it from any node with no predecessor.
- *
- * Longest path rather than shortest, because a node has to come after EVERYTHING that feeds
- * it. Take the shortest and an arrow ends up pointing backwards: with `a→b`, `a→c`, `c→b`,
- * the shortest path puts b in layer 1 alongside c, and then `c→b` points sideways.
- */
-function assignLayers(nodes: string[], forward: GraphEdge[]): string[][] {
-    const layer = new Map<string, number>(nodes.map((n) => [n, 0]))
-    // Relaxation, bounded by the node count: the longest possible chain visits every node
-    // once, so after that many rounds nothing can still be moving.
-    for (let round = 0; round < nodes.length; round++) {
-        let moved = false
-        for (const e of forward) {
-            const want = (layer.get(e.source) ?? 0) + 1
-            if (want > (layer.get(e.target) ?? 0)) {
-                layer.set(e.target, want)
-                moved = true
-            }
-        }
-        if (!moved) break
-    }
-    const depth = Math.max(0, ...layer.values()) + 1
-    const layers: string[][] = Array.from({ length: depth }, () => [])
-    // Declaration order within a layer, so the ordering pass starts somewhere predictable.
-    for (const n of nodes) layers[layer.get(n) ?? 0].push(n)
-    return layers
-}
-
-/**
- * Reorder each layer to reduce the number of arrows that cross.
- *
- * Barycentre sweeping: a node is placed at the average position of the nodes it connects to
- * in the neighbouring layer, and the whole diagram is swept downwards then upwards
- * repeatedly. Each sweep can only be judged against the previous layer's order, so a node
- * pulled into a better place drags its own neighbours in the next sweep.
- *
- * The heuristic, not an exact minimum: finding the true minimum number of crossings is
- * NP-hard even for two layers. In practice this reaches zero crossings on the flowcharts the
- * model actually produces — verified on a 14-node pipeline with two diamonds and a rollback
- * loop, and on a bipartite graph whose declared order forces three crossings.
- */
-function reduceCrossings(layers: string[][], edges: GraphEdge[]): void {
-    if (layers.length < 2) return
-    const PASSES = 8
-    const into = new Map<string, string[]>()
-    const outOf = new Map<string, string[]>()
-    for (const e of edges) {
-        if (e.source === e.target) continue
-        ;(into.get(e.target) ?? into.set(e.target, []).get(e.target))?.push(
-            e.source,
-        )
-        ;(outOf.get(e.source) ?? outOf.set(e.source, []).get(e.source))?.push(
-            e.target,
-        )
-    }
-
-    let best = layers.map((l) => [...l])
-    let bestScore = countCrossings(layers, edges)
-
-    for (let pass = 0; pass < PASSES && bestScore > 0; pass++) {
-        const pos = new Map<string, number>()
-        for (const l of layers)
-            l.forEach((n, i) => {
-                pos.set(n, i)
-            })
-        const down = pass % 2 === 0
-        const order = down
-            ? layers.map((_, i) => i).slice(1)
-            : layers
-                  .map((_, i) => i)
-                  .slice(0, -1)
-                  .reverse()
-
-        for (const i of order) {
-            const neighbours = down ? into : outOf
-            const key = new Map<string, number>()
-            layers[i].forEach((n, idx) => {
-                const nb = (neighbours.get(n) ?? [])
-                    .map((m) => pos.get(m))
-                    .filter((v): v is number => v !== undefined)
-                // A node with no neighbour in that direction keeps its place, rather than
-                // being pushed to one end by a default of zero.
-                key.set(
-                    n,
-                    nb.length ? nb.reduce((a, b) => a + b, 0) / nb.length : idx,
-                )
-            })
-            layers[i] = [...layers[i]].sort(
-                (a, b) => (key.get(a) ?? 0) - (key.get(b) ?? 0),
-            )
-        }
-
-        // Keep the best arrangement seen: sweeping is not monotonic, and a later pass can be
-        // worse than an earlier one.
-        const score = countCrossings(layers, edges)
-        if (score < bestScore) {
-            bestScore = score
-            best = layers.map((l) => [...l])
-        }
-    }
-    for (let i = 0; i < layers.length; i++) layers[i] = best[i]
-}
-
-/**
- * How many pairs of arrows cross between adjacent layers.
- *
- * Two arrows between the same pair of layers cross exactly when their endpoints are in the
- * opposite order on the two sides. That is all this counts — arrows spanning more than one
- * layer are ignored here, because their crossings depend on routing rather than ordering.
- */
-function countCrossings(layers: string[][], edges: GraphEdge[]): number {
-    const layerOf = new Map<string, number>()
-    const posOf = new Map<string, number>()
-    layers.forEach((l, i) => {
-        l.forEach((n, j) => {
-            layerOf.set(n, i)
-            posOf.set(n, j)
-        })
-    })
-    let total = 0
-    for (let i = 0; i + 1 < layers.length; i++) {
-        const span = edges.filter(
-            (e) =>
-                layerOf.get(e.source) === i && layerOf.get(e.target) === i + 1,
-        )
-        for (let a = 0; a < span.length; a++)
-            for (let b = a + 1; b < span.length; b++) {
-                const s1 = posOf.get(span[a].source) ?? 0
-                const t1 = posOf.get(span[a].target) ?? 0
-                const s2 = posOf.get(span[b].source) ?? 0
-                const t2 = posOf.get(span[b].target) ?? 0
-                if ((s1 - s2) * (t1 - t2) < 0) total++
-            }
-    }
-    return total
-}
-
-/**
- * Turn a graph into the operations that draw it.
- *
- * The output is ordinary operations — nothing here is a new kind of thing the rest of the
- * engine has to know about. A layer of one node is emitted directly rather than wrapped,
- * because a single-child row container would just add a level of nesting with nothing to
- * arrange.
- */
-export function graphToOperations(
-    nodes: GraphNode[],
-    edges: GraphEdge[],
-    opts: GraphOptions = {},
-): GraphResult {
-    const flow = opts.flow ?? "col"
-    const ids = nodes.map((n) => n.id)
-    const known = new Set(ids)
-
-    const unknownEndpoints: string[] = []
-    const usable: GraphEdge[] = []
-    for (const e of edges) {
-        if (!known.has(e.source)) unknownEndpoints.push(e.source)
-        if (!known.has(e.target)) unknownEndpoints.push(e.target)
-        if (known.has(e.source) && known.has(e.target)) usable.push(e)
-    }
-
-    // A self-loop tells us nothing about layering and would make the cycle break drop a real
-    // arrow, so it is set aside and drawn as-is.
-    const loops = usable.filter((e) => e.source === e.target)
-    const between = usable.filter((e) => e.source !== e.target)
-
-    const { forward, back } = breakCycles(ids, between)
-    const layers = assignLayers(ids, forward)
-    reduceCrossings(layers, forward)
-
-    // The flow axis is the OUTER container's direction; a layer runs across it.
-    const outerDir = flow
-    const layerDir = flow === "col" ? "row" : "col"
-    const ns = opts.prefix ?? ""
-    const root = opts.rootId ?? `${ns}${LAYER_ID}s`
-
-    const operations: Operation[] = [
-        {
-            op: "add_container",
-            id: root,
-            ...(opts.parent ? { parent: opts.parent } : {}),
-            label: "",
-            dir: outerDir,
-            gap: LAYER_GAP,
-        },
-    ]
-    const byId = new Map(nodes.map((n) => [n.id, n]))
-    const add = (id: string, parent: string): Operation => {
-        const n = byId.get(id) as GraphNode
-        return n.icon
-            ? {
-                  op: "add_icon",
-                  id: n.id,
-                  parent,
-                  name: n.icon,
-                  label: n.label,
-              }
-            : {
-                  op: "add_box",
-                  id: n.id,
-                  parent,
-                  label: n.label,
-                  ...(n.shape && n.shape !== "box" ? { shape: n.shape } : {}),
-                  ...(n.role && n.role !== "body" ? { role: n.role } : {}),
-                  ...(n.group ? { group: n.group } : {}),
-              }
-    }
-
-    layers.forEach((members, i) => {
-        if (members.length === 0) return
-        if (members.length === 1) {
-            operations.push(add(members[0], root))
-            return
-        }
-        const band = `${ns}${LAYER_ID}${i}`
-        operations.push({
-            op: "add_container",
-            id: band,
-            parent: root,
-            label: "",
-            dir: layerDir,
-            gap: NODE_GAP,
-        })
-        for (const m of members) operations.push(add(m, band))
-    })
-
-    for (const e of [...between, ...loops])
-        operations.push({
-            op: "link",
-            source: e.source,
-            target: e.target,
-            ...(e.label ? { label: e.label } : {}),
-            ...(e.dashed ? { dashed: true } : {}),
-            ...(e.bold ? { bold: true } : {}),
-            ...(e.head !== undefined
-                ? { head: e.head, headFill: e.headFill ?? false }
-                : {}),
-            ...(e.tail !== undefined
-                ? { tail: e.tail, tailFill: e.tailFill ?? false }
-                : {}),
-        })
-
-    return {
-        operations,
-        layers: layers.filter((l) => l.length > 0),
-        unknownEndpoints: [...new Set(unknownEndpoints)],
-        backEdges: back.map((e) => ({ source: e.source, target: e.target })),
-    }
-}

lib/diagram-engine/index.ts

@@ -1,146 +0,0 @@
-/**
- * The engine's entry point: one call takes the current canvas XML plus a list of
- * structural operations and returns new canvas XML.
- *
- *   current XML → parse → apply operations → check names → layout → render → new XML
- *
- * The tree is not stored anywhere between calls. It is re-derived from the canvas every
- * time, so a user's manual edits — moving a shape into a different frame, recolouring a
- * box, adding an annotation — are simply part of the input to the next layout. There is
- * no second copy of the state, and therefore nothing to reconcile.
- */
-
-import { checkNames, resolveStyle } from "./catalog"
-import {
-    applyOperations,
-    collectNames,
-    type Operation,
-    outline,
-} from "./operations"
-import { parseDiagram } from "./parse"
-import { renderDiagram } from "./render"
-import { nearestShape, resolveShape } from "./shapes"
-import { type DiagramTree, walkTree } from "./types"
-
-export interface RestructureResult {
-    /** New canvas XML, or null when the request could not be carried out. */
-    xml: string | null
-    /** Compact outline of the resulting structure, for the model to read back. */
-    outline: string
-    /** Operations that could not be applied, and invented stencil names. */
-    errors: string[]
-    /** Non-fatal notes: pages skipped, structure that could not be read cleanly. */
-    warnings: string[]
-}
-
-export interface RestructureOptions {
-    /** Which page of a multi-page document to work on. */
-    pageIndex?: number
-    /** Diagram-wide icon glyph size. */
-    iconSize?: number
-}
-
-/**
- * Apply structural operations to whatever is on the canvas.
- *
- * `currentXml` may be empty — that is how a diagram gets built from scratch.
- *
- * An invented stencil name is a hard error, not a silent fallback: draw.io renders an
- * unknown `resIcon` as a blank square, so a diagram that "worked" would be quietly
- * missing icons. The error carries suggestions from the catalog so the model can fix it
- * in one more turn.
- */
-export function restructureDiagram(
-    currentXml: string,
-    ops: Operation[],
-    opts: RestructureOptions = {},
-): RestructureResult {
-    const warnings: string[] = []
-
-    let tree: DiagramTree
-    if (currentXml.trim()) {
-        const parsed = parseDiagram(currentXml, opts.pageIndex ?? 0)
-        tree = parsed.tree
-        warnings.push(...parsed.warnings)
-    } else {
-        tree = { roots: [], links: [], foreign: [] }
-    }
-
-    const applied = applyOperations(tree, ops)
-    const errors = [...applied.errors]
-    warnings.push(...applied.warnings)
-
-    // Catch invented names before rendering, so the model gets a correctable error
-    // instead of a diagram with blank squares in it.
-    for (const bad of checkNames(collectNames(applied.tree))) {
-        const hint = bad.suggestions.length
-            ? ` Did you mean: ${bad.suggestions.join(", ")}?`
-            : ""
-        errors.push(
-            `"${bad.name}" (node ${bad.id}) is not in the stencil catalog.${hint}`,
-        )
-    }
-
-    // Shape tokens: an injection-capable token is an error; an unknown-but-safe one
-    // passes through (draw.io degrades it to a rectangle) but gets a warning, so a typo
-    // is a one-turn fix instead of a silently rectangular "cyclinder" forever.
-    for (const n of walkTree(applied.tree)) {
-        if (n.kind !== "box" || !n.shape || n.shape === "box") continue
-        const resolved = resolveShape(n.shape)
-        if (!resolved) {
-            errors.push(
-                `shape "${n.shape}" (node ${n.id}) contains characters that are not allowed in a shape token.`,
-            )
-        } else if (resolved.passthrough) {
-            const near = nearestShape(n.shape)
-            warnings.push(
-                `shape "${n.shape}" (node ${n.id}) is not in the engine's catalog — passed through to draw.io, which renders unknown shapes as rectangles.${near ? ` Did you mean "${near}"?` : ""}`,
-            )
-        }
-    }
-
-    if (errors.length > 0)
-        return { xml: null, outline: outline(applied.tree), errors, warnings }
-
-    const rendered = renderDiagram(applied.tree, {
-        resolveStyle,
-        iconSize: opts.iconSize,
-    })
-    if (rendered.danglingLinks.length)
-        warnings.push(
-            `Dropped edge(s) pointing at missing nodes: ${rendered.danglingLinks.join(", ")}.`,
-        )
-
-    return {
-        xml: rendered.xml,
-        outline: outline(applied.tree),
-        errors: [],
-        warnings,
-    }
-}
-
-/** Read the current canvas structure without changing it. */
-export function describeDiagram(
-    currentXml: string,
-    pageIndex = 0,
-): { outline: string; warnings: string[]; needsAdoption: boolean } {
-    if (!currentXml.trim())
-        return { outline: "(empty canvas)", warnings: [], needsAdoption: false }
-    const { tree, warnings, needsAdoption } = parseDiagram(
-        currentXml,
-        pageIndex,
-    )
-    return { outline: outline(tree), warnings, needsAdoption }
-}
-
-export { CATALOG_SIZE, lookupStencil, searchStencils } from "./catalog"
-export {
-    type GraphEdge,
-    type GraphNode,
-    type GraphOptions,
-    graphToOperations,
-} from "./graph"
-export { type Operation, OperationSchema } from "./operations"
-export { parseDiagram } from "./parse"
-export { renderDiagram } from "./render"
-export type { DiagramNode, DiagramTree } from "./types"

lib/diagram-engine/markers.ts

@@ -1,585 +0,0 @@
-/**
- * Style markers — how layout structure survives a round-trip through draw.io.
- *
- * The layout engine's tree carries information plain draw.io XML does not: which
- * direction a container stacks its children, the gap between them, and whether the
- * user has pinned a node's position. We encode that as extra `key=value` tokens in
- * the cell's style string.
- *
- * Two behaviours this relies on, both verified in a real browser (Playwright drag
- * against the embedded editor, reading the editor's own autosave payload):
- *
- *   1. draw.io PRESERVES style keys it does not understand. After a user drags a
- *      shape and the editor saves, `dai_kind=group;dai_dir=col;dai_gap=22;` came
- *      back byte-identical.
- *   2. On a DUPLICATE key, the LAST value wins. A style ending in
- *      `container=0;pointerEvents=0;container=1;` behaved as a container: a shape
- *      dragged into it was reparented. So we can append a normalising token without
- *      first parsing out the old one.
- *
- * (2) matters because the AWS catalog is inconsistent: group_region, group_vpc,
- * group_subnet, group_availability_zone, group_aws_cloud and group_on_premise ship
- * WITHOUT container=1, while group_account, group_aws_cloud_alt, group_vpc2,
- * group_security_group and group_corporate_data_center ship WITH it. Appending
- * unconditionally normalises all of them.
- */
-
-/** Marker keys. Namespaced with `dai_` so they cannot collide with mxGraph keys. */
-export const MARKER = {
-    /** Node kind, so the parser does not have to re-guess it from the shape. */
-    kind: "dai_kind",
-    /** Child stacking direction of a container: "row" | "col" | "grid". */
-    dir: "dai_dir",
-    /** Gap between children, in px. */
-    gap: "dai_gap",
-    /** Column count, for grid containers. */
-    cols: "dai_cols",
-    /** Set by the user to freeze a node's position across re-layouts. */
-    pin: "dai_pin",
-    /**
-     * A catalog icon's name. Needed because an Azure or GCP icon's style is an embedded
-     * base64 image with no name anywhere in it, so the style alone cannot identify it.
-     */
-    name: "dai_name",
-    /**
-     * Which (lane, column) cell of a swimlane pool a node occupies, as "lane,col".
-     *
-     * Position alone cannot recover this once the user drags a node: the cell it lands in
-     * is a guess, whereas the marker records which lane the model assigned it to. It is
-     * also the only way an empty cell stays empty — geometry can only tell us where things
-     * ARE, never that a role deliberately does nothing at a given step.
-     */
-    cell: "dai_cell",
-    /** A pool's lane names, tab-separated (a tab cannot appear in a draw.io style value). */
-    lanes: "dai_lanes",
-    /** A pool's milestone labels, tab-separated. */
-    phases: "dai_phases",
-    /** A pool's orientation: "h" or "v". */
-    orient: "dai_orient",
-    /** Vertical distance between consecutive messages in a sequence diagram. */
-    step: "dai_step",
-    /** How a radial container fans its branches out: "radial" or "down". */
-    spread: "dai_spread",
-    /** The node's information role (banner, heading, callout…), for the round trip. */
-    role: "dai_role",
-    /** The node's semantic zone, whose hue ramp colours it. */
-    group: "dai_group",
-    /**
-     * The declared shape token, verbatim. Appearance-based reverse mapping is ambiguous
-     * (aliases, rotated variants, styles with no unique shape= token), so the round trip
-     * carries the declaration itself.
-     */
-    shape: "dai_shape",
-    /**
-     * Marks a node's size as engine-computed rather than user-fixed. Without it, the
-     * w/h read back from the canvas would freeze the first layout's measurement: change
-     * the label and the box would keep the old size instead of re-measuring.
-     */
-    auto: "dai_auto",
-    /** Share of the parent's leftover flow-axis space — flex-grow. */
-    grow: "dai_grow",
-    /** Cross-axis position within the parent: "start" | "center" | "end". */
-    align: "dai_align",
-    /** How a container spreads children along its own axis — justify-content. */
-    justify: "dai_justify",
-    /** A container's cross-axis default for children that declare no align of their own. */
-    alignItems: "dai_aitems",
-    /** Opted out of the content-width floor when weights divide a row — CSS's min-width:0. */
-    minw0: "dai_minw0",
-    /**
-     * Declared width cap, px.
-     *
-     * Has to be a marker rather than inferred from the drawn width: the two are only equal
-     * when the cap actually bit. A box capped at 400 that happens to be 260 wide would come
-     * back with a 260 cap, and the next re-layout could never let it grow again.
-     */
-    maxw: "dai_maxw",
-    /** A container's interior padding, px. */
-    pad: "dai_pad",
-    /**
-     * The page's declared width:height, on the default layer's cell.
-     *
-     * Page-level rather than per-node, so it goes on layer "1" — the one cell every
-     * diagram has and draw.io never discards. It cannot be inferred from pageWidth and
-     * pageHeight: those are what the last layout produced, so reading them back would
-     * turn whatever shape a diagram happened to come out as into a standing request to
-     * keep it.
-     */
-    aspect: "dai_aspect",
-    /**
-     * Marks a cell as chrome the engine draws and owns: a pool's lane bands, its label
-     * columns, its milestone strip. The parser must not read these back as nodes — they are
-     * re-derived from the pool's own parameters on every layout — and the edge router must
-     * not treat them as obstacles, since a sequence flow crossing lanes is the norm.
-     */
-    lane: "dai_lane",
-} as const
-
-export type NodeKind =
-    | "group"
-    | "grid"
-    | "pool"
-    | "sequence"
-    | "radial"
-    | "icon"
-    | "box"
-    | "title"
-export type Direction = "row" | "col" | "grid"
-
-/**
- * Tokens that make a shape behave as a container in draw.io: it accepts a shape
- * dragged into it and reparents that shape (setting `parent` and switching the
- * child's geometry to parent-relative).
- *
- * `pointerEvents=0` keeps clicks falling through to the children — without it the
- * frame swallows them and the user cannot select what is inside. `collapsible=0`
- * hides the fold arrow. `recursiveResize=0` stops children from being scaled when
- * the frame is resized, which would fight the layout engine.
- */
-const CONTAINER_TOKENS =
-    "container=1;pointerEvents=0;collapsible=0;recursiveResize=0;"
-
-/**
- * A container that groups children for layout but should not be visible.
- *
- * The reference project solves this with a "phantom": a wrapper that participates in
- * layout and then emits NO cell, reparenting its children onto the nearest visible
- * ancestor. That makes the round-trip lossy by construction — the wrapper's direction
- * and grouping are simply absent from the XML, so re-deriving the tree cannot recover
- * them. Measured on the reference project's own build_vpc.mjs: a phantom erased a
- * container's "col" direction, leaving children in a 2-D arrangement that can only be
- * read back as a grid.
- *
- * So we emit a real cell and make it invisible instead. One extra cell per wrapper,
- * in exchange for structure that survives being read back.
- */
-const INVISIBLE_TOKENS = "fillColor=none;strokeColor=none;"
-
-/** Read a marker's raw value out of a style string. Last occurrence wins, as draw.io does. */
-export function readMarker(style: string, key: string): string | null {
-    // Scan all matches and keep the last, mirroring draw.io's duplicate-key resolution.
-    const re = new RegExp(`(?:^|;)${key}=([^;]*)`, "g")
-    let last: string | null = null
-    let m = re.exec(style)
-    while (m !== null) {
-        last = m[1]
-        m = re.exec(style)
-    }
-    return last
-}
-
-const KINDS: readonly NodeKind[] = [
-    "group",
-    "grid",
-    "pool",
-    "sequence",
-    "radial",
-    "icon",
-    "box",
-    "title",
-]
-
-export function readKind(style: string): NodeKind | null {
-    const v = readMarker(style, MARKER.kind)
-    return KINDS.includes(v as NodeKind) ? (v as NodeKind) : null
-}
-
-/**
- * The (lane, column) cell a node occupies in a swimlane pool, or null.
- *
- * Both must be non-negative integers: a malformed value is safer read as "no cell
- * declared" (which puts the node in lane 0 column 0) than as a negative index, which would
- * place it outside the pool's frame.
- */
-export function readCell(style: string): { lane: number; col: number } | null {
-    const v = readMarker(style, MARKER.cell)
-    if (!v) return null
-    const m = v.match(/^(\d+),(\d+)$/)
-    return m ? { lane: Number(m[1]), col: Number(m[2]) } : null
-}
-
-/**
- * A tab-separated marker list, as written by `joinList`.
- *
- * A tab cannot appear in a draw.io style value — the editor writes styles as a single
- * semicolon-separated line — so it is safe as a separator inside one value, where a comma
- * would collide with the label text it has to carry.
- */
-export function readList(style: string, key: string): string[] | null {
-    const v = readMarker(style, key)
-    if (v === null) return null
-    if (v === "") return []
-    return v.split("\t").map(decodeURIComponent)
-}
-
-/** Encode a list of labels into one marker value. */
-export function joinList(items: string[]): string {
-    // Percent-encoding keeps a label containing ";" or "=" from breaking the style string.
-    return items.map((s) => encodeURIComponent(s)).join("\t")
-}
-
-/** Is this cell pool chrome the engine draws and owns, rather than a node? */
-export function isLaneChrome(style: string): boolean {
-    return readMarker(style, MARKER.lane) !== null
-}
-
-export function readDir(style: string): Direction | null {
-    const v = readMarker(style, MARKER.dir)
-    if (v === "row" || v === "col" || v === "grid") return v
-    return null
-}
-
-/** Read a positive integer marker (gap, cols). Returns null when absent or malformed. */
-export function readIntMarker(style: string, key: string): number | null {
-    const v = readMarker(style, key)
-    if (v === null) return null
-    const n = Number(v)
-    return Number.isFinite(n) && n >= 0 ? Math.round(n) : null
-}
-
-/**
- * Has the user pinned this node? Any value other than "0"/""/"false" counts as
- * pinned, so a user typing `dai_pin=1` (or just `dai_pin=yes`) in draw.io's
- * "Edit Style" dialog gets what they expect.
- */
-export function isPinned(style: string): boolean {
-    const v = readMarker(style, MARKER.pin)
-    if (v === null) return false
-    const s = v.trim().toLowerCase()
-    return s !== "" && s !== "0" && s !== "false"
-}
-
-/**
- * Append `key=value;`, replacing any existing occurrence of the key first.
- *
- * Styles are re-stamped on every render, and a style recovered from the canvas already
- * carries last render's markers — blindly appending grew the string by one duplicate per
- * round-trip, unboundedly. Duplicates resolve last-wins in draw.io so nothing ever LOOKED
- * wrong, which is why it went unnoticed until a byte-identity test caught it.
- *
- * Only `dai_*` keys are cleaned. mxGraph keys are appended verbatim because last-wins is
- * load-bearing there: the container tokens rely on appending `container=1` after a catalog
- * style that may say `container=0`.
- */
-/**
- * Set each `key=value;` token of `tokens` on a style, replacing any value already there.
- *
- * Matching is per token, not on the whole run: a catalog style may already declare
- * `container=1` while saying nothing about `pointerEvents`, and re-adding the whole run
- * because one token was missing is what let these accumulate.
- *
- * Exported because the same defect appeared a second time, on EDGES: an edge's style starts
- * from whatever the canvas held, which already carried the previous pass's `exitX`/`entryX`
- * port keys, and the router appended a fresh set on top of them every render — 76 characters
- * per round-trip, without bound. Any code that re-stamps a computed mxGraph key onto a style
- * recovered from the canvas needs this rather than `+=`.
- */
-export function appendOnce(style: string, tokens: string): string {
-    let s = style
-    for (const tok of tokens.split(";")) {
-        if (!tok) continue
-        const key = tok.slice(0, tok.indexOf("="))
-        // The key must not be present with ANY value: `container=0` from a catalog stencil
-        // has to be overwritten, which is what appending the correct value does.
-        const has = new RegExp(`(?:^|;)${key}=[^;]*;`).test(s)
-        if (has) {
-            s = s.replace(new RegExp(`(?:^|(?<=;))${key}=[^;]*;`, "g"), "")
-        }
-        s = s.endsWith(";") || s === "" ? s : `${s};`
-        s += `${tok};`
-    }
-    return s
-}
-
-function append(style: string, key: string, value: string | number): string {
-    const cleaned = key.startsWith("dai_")
-        ? style.replace(new RegExp(`(?:^|(?<=;))${key}=[^;]*;`, "g"), "")
-        : style
-    const base =
-        cleaned.endsWith(";") || cleaned === "" ? cleaned : `${cleaned};`
-    return `${base}${key}=${value};`
-}
-
-/**
- * Stamp a container's style: make it a real draw.io container and record its
- * layout parameters.
- *
- * Appends rather than rewrites. Duplicate keys are legal and the last one wins, so
- * a catalog style that already says `container=1` is unharmed, and one that says
- * nothing (or `container=0`) is corrected.
- */
-export function stampContainer(
-    style: string,
-    opts: {
-        kind: "group" | "grid"
-        dir: Direction
-        gap: number
-        cols?: number
-        /** Layout-only wrapper: emit a real cell, but draw nothing. */
-        invisible?: boolean
-    },
-): string {
-    let s = style.endsWith(";") || style === "" ? style : `${style};`
-    // Appended only when not already there. These are plain mxGraph keys, so `append`'s
-    // de-duplication (which is limited to `dai_*`) does not cover them — and a container
-    // goes through here on EVERY re-layout, so a blind `+=` grew the style string by
-    // another `container=1;pointerEvents=0;collapsible=0;recursiveResize=0;` per round
-    // trip, without bound. Harmless to draw.io, which takes the last value, but the XML
-    // never reached a fixed point and every edit shipped a longer style.
-    s = appendOnce(s, CONTAINER_TOKENS)
-    if (opts.invisible) s = appendOnce(s, INVISIBLE_TOKENS)
-    s = append(s, MARKER.kind, opts.kind)
-    s = append(s, MARKER.dir, opts.dir)
-    s = append(s, MARKER.gap, Math.round(opts.gap))
-    if (opts.kind === "grid" && opts.cols != null)
-        s = append(s, MARKER.cols, Math.max(1, Math.round(opts.cols)))
-    return s
-}
-
-/**
- * Stamp a swimlane pool: its lane names, milestone labels and orientation.
- *
- * Unlike a group, a pool is NOT stamped as a draw.io container. Its lane bands are separate
- * cells sitting inside it, and they are what a shape should reparent into when the user
- * drags it — that is how "the user moved this step to a different role" gets recorded. If
- * the pool itself claimed the drop, every node would come back in lane 0.
- */
-export function stampPool(
-    style: string,
-    opts: {
-        lanes: string[]
-        phases: string[]
-        orientation: "horizontal" | "vertical"
-        gap: number
-    },
-): string {
-    let s = append(style, MARKER.kind, "pool")
-    s = append(s, MARKER.lanes, joinList(opts.lanes))
-    s = append(s, MARKER.phases, joinList(opts.phases))
-    s = append(s, MARKER.orient, opts.orientation === "vertical" ? "v" : "h")
-    return append(s, MARKER.gap, Math.round(opts.gap))
-}
-
-/** Stamp a sequence container: participant spacing and message spacing. */
-export function stampSequence(
-    style: string,
-    opts: { gap: number; step: number },
-): string {
-    const s = append(style, MARKER.kind, "sequence")
-    return append(
-        append(s, MARKER.gap, Math.round(opts.gap)),
-        MARKER.step,
-        Math.round(opts.step),
-    )
-}
-
-/** Stamp a radial container: how it fans branches out, and the ring spacing. */
-export function stampRadial(
-    style: string,
-    opts: { spread: "radial" | "down"; gap: number },
-): string {
-    const s = append(style, MARKER.kind, "radial")
-    return append(
-        append(s, MARKER.spread, opts.spread),
-        MARKER.gap,
-        Math.round(opts.gap),
-    )
-}
-
-/**
- * Stamp one of a pool's lane bands.
- *
- * A band IS a draw.io container, so dragging a step onto another role's band reparents it
- * there and the marker on the band tells the parser which lane that is. The lane index is
- * the band's identity, not its position, so the assignment survives the pool being
- * re-measured to a different size.
- */
-export function stampLane(style: string, lane: number): string {
-    let s = style.endsWith(";") || style === "" ? style : `${style};`
-    s += CONTAINER_TOKENS
-    return append(s, MARKER.lane, Math.max(0, Math.round(lane)))
-}
-
-/**
- * Stamp a pool's own decoration — a lane-name column or a milestone strip.
- *
- * `dai_lane=-1` marks it as chrome the renderer rebuilds, so the parser drops it rather
- * than reading it back as a node. Unlike a lane band it is deliberately NOT a draw.io
- * container: a step dropped on a label column belongs to no role, and letting it reparent
- * there would lose the step's lane.
- */
-export function stampPoolDecoration(style: string): string {
-    return append(style, MARKER.lane, -1)
-}
-
-/** Record which pool cell a node occupies. */
-export function stampCell(
-    style: string,
-    cell: { lane: number; col: number },
-): string {
-    return append(
-        style,
-        MARKER.cell,
-        `${Math.max(0, Math.round(cell.lane))},${Math.max(0, Math.round(cell.col))}`,
-    )
-}
-
-/**
- * Is this an invisible layout wrapper? Both colours set to `none` and no group
- * stencil — a visible frame always has a stroke or a stencil.
- */
-export function isInvisible(style: string): boolean {
-    if (/grIcon=/.test(style)) return false
-    const fill = readMarker(style, "fillColor")
-    const stroke = readMarker(style, "strokeColor")
-    return fill === "none" && stroke === "none"
-}
-
-/**
- * Stamp a leaf with its kind, so the parser need not infer it.
- *
- * For an icon, also record the catalog name: an Azure or GCP icon's style is an embedded
- * base64 image with no name in it, so the style alone cannot identify which icon it is.
- */
-export function stampLeaf(
-    style: string,
-    kind: "icon" | "box" | "title",
-    opts: { name?: string } = {},
-): string {
-    const s = append(style, MARKER.kind, kind)
-    return opts.name ? append(s, MARKER.name, opts.name) : s
-}
-
-/** Stamp the node's information role, replacing any previous one. */
-export function stampRole(style: string, role: string): string {
-    return append(style, MARKER.role, role)
-}
-
-/** Stamp the node's semantic zone, replacing any previous one. */
-export function stampGroup(style: string, group: string): string {
-    return append(style, MARKER.group, encodeURIComponent(group))
-}
-
-/** Stamp the declared shape token, so the round trip carries the declaration itself. */
-export function stampShape(style: string, shape: string): string {
-    return append(style, MARKER.shape, encodeURIComponent(shape))
-}
-
-/** Mark a node's size as engine-computed, so a re-layout re-measures it. */
-export function stampAuto(style: string): string {
-    return append(style, MARKER.auto, 1)
-}
-
-/** Was this node's size computed by the engine (vs fixed by the user or the model)? */
-export function isAutoSized(style: string): boolean {
-    return readMarker(style, MARKER.auto) === "1"
-}
-
-type FlexAlign = "start" | "center" | "end" | "stretch"
-type FlexJustify = "start" | "center" | "end" | "between" | "around" | "evenly"
-
-/** Stamp the flex fields a node carries, so a round-trip preserves them. */
-export function stampFlex(
-    style: string,
-    opts: {
-        grow?: number
-        align?: FlexAlign
-        justify?: FlexJustify
-        alignItems?: FlexAlign
-        maxW?: number
-        minW0?: boolean
-        pad?: number
-    },
-): string {
-    let s = style
-    if (opts.grow != null && opts.grow > 0)
-        s = append(s, MARKER.grow, opts.grow)
-    if (opts.align && opts.align !== "center")
-        s = append(s, MARKER.align, opts.align)
-    if (opts.justify && opts.justify !== "start")
-        s = append(s, MARKER.justify, opts.justify)
-    if (opts.alignItems) s = append(s, MARKER.alignItems, opts.alignItems)
-    if (opts.maxW != null && opts.maxW > 0)
-        s = append(s, MARKER.maxw, Math.round(opts.maxW))
-    if (opts.minW0) s = append(s, MARKER.minw0, 1)
-    if (opts.pad != null) s = append(s, MARKER.pad, Math.round(opts.pad))
-    return s
-}
-
-/** Read the align marker back. Anything unrecognised means the default (center). */
-export function readAlign(style: string): Exclude<FlexAlign, "center"> | null {
-    const v = readMarker(style, MARKER.align)
-    return v === "start" || v === "end" || v === "stretch" ? v : null
-}
-
-/** Read a container's cross-axis default. Null means it declared none. */
-export function readAlignItems(style: string): FlexAlign | null {
-    const v = readMarker(style, MARKER.alignItems)
-    return v === "start" || v === "end" || v === "stretch" || v === "center"
-        ? v
-        : null
-}
-
-/** Read the justify marker back. Anything unrecognised means the default (start). */
-export function readJustify(
-    style: string,
-): Exclude<FlexJustify, "start"> | null {
-    const v = readMarker(style, MARKER.justify)
-    return v === "center" ||
-        v === "end" ||
-        v === "between" ||
-        v === "around" ||
-        v === "evenly"
-        ? v
-        : null
-}
-
-/** Read the declared width cap back, or null when there was none. */
-export function readMaxW(style: string): number | null {
-    const v = Number(readMarker(style, MARKER.maxw))
-    return Number.isFinite(v) && v > 0 ? v : null
-}
-
-/** Did this node opt out of the content-width floor? */
-export function readMinW0(style: string): boolean {
-    return readMarker(style, MARKER.minw0) === "1"
-}
-
-/** The page's declared aspect ratio, stamped on the default layer. */
-export function stampAspect(layerXml: string, aspect: number): string {
-    return layerXml.replace(
-        /<mxCell id="1" parent="0"\/>/,
-        `<mxCell id="1" parent="0" style="${MARKER.aspect}=${aspect};"/>`,
-    )
-}
-
-/**
- * Read the page's declared aspect back out of a model body.
- *
- * Scans for the marker anywhere in the page rather than parsing the layer cell: the
- * marker name is namespaced, so a match cannot be anything else, and this keeps working
- * if draw.io ever reorders or reformats that cell.
- */
-export function readAspect(page: string): number | undefined {
-    const m = new RegExp(`${MARKER.aspect}=([\\d.]+)`).exec(page)
-    if (!m) return undefined
-    const v = Number(m[1])
-    return Number.isFinite(v) && v > 0
-        ? Math.min(4, Math.max(0.25, v))
-        : undefined
-}
-
-/** Strip every `dai_*` marker — for exporting a clean file, or comparing styles. */
-export function stripMarkers(style: string): string {
-    return style
-        .split(";")
-        .filter((tok) => tok !== "" && !tok.startsWith("dai_"))
-        .join(";")
-        .concat(";")
-        .replace(/^;$/, "")
-}
-
-/** Does this style carry any engine marker? Used to tell engine output from imported files. */
-export function hasMarkers(style: string): boolean {
-    return /(?:^|;)dai_[a-z]+=/.test(style)
-}

lib/diagram-engine/shapes.ts

@@ -1,235 +0,0 @@
-/**
- * The shape vocabulary: what a box can BE, beyond a labelled rectangle.
- *
- * draw.io ships hundreds of shapes; the engine's declarative layer used to allow six.
- * That gap — not colours, not spacing — was why engine output looked flat next to
- * hand-written XML: a database drawn as a grey rectangle labelled "database" instead of
- * a cylinder. This module opens the vocabulary in two tiers:
- *
- *   CATALOG — ~20 curated shapes the engine fully understands. Each entry carries the
- *   complete style fragment (including the matching `perimeter=`, which draw.io's own
- *   style reference warns is required or edges connect to the bounding box), how much
- *   larger the box must be for its text to fit inside the non-rectangular outline
- *   (verified empirically in the real editor: a rhombus needs ~1.5× the rectangle's
- *   size for the same text), and whether the label renders below the glyph instead of
- *   inside it.
- *
- *   PASS-THROUGH — any other token that looks like a draw.io shape name is emitted
- *   verbatim as `shape=<token>;`. Verified in the real editor: an unknown token
- *   degrades to a rectangle, it does not break the page. A conservative text scale
- *   covers the common case that the real shape is roughly convex. The tool response
- *   carries a near-match hint ("cyclinder → cylinder?") so a typo is a one-turn fix,
- *   not a silent permanent degradation.
- *
- * Style strings are merged structurally, not concatenated: each fragment is parsed to
- * key=value tokens and later fragments override earlier ones per key. This is what
- * makes shape and theme composable by rule — the shape fragment owns geometry keys
- * (shape, perimeter, rounded…), the theme owns colour and type keys, and an overlap
- * (a theme that says rounded=1 on a rhombus) resolves by order instead of emitting
- * two conflicting tokens.
- */
-
-/** How a known shape renders and measures. */
-export interface ShapeSpec {
-    /** Geometry style tokens ONLY — no colours, no fonts; those belong to the theme. */
-    style: string
-    /**
-     * How much larger than a rectangle the box must be for the same text to fit
-     * inside the outline. 1.0 for the rectangle family; ~1.5 for a rhombus, whose
-     * inscribed rectangle is half its bounding box.
-     */
-    textScale: number
-    /** The label renders below the glyph, not inside it (umlActor and friends). */
-    labelOutside?: boolean
-    /** Fixed glyph size for labelOutside shapes, which do not scale with text. */
-    glyph?: { w: number; h: number }
-}
-
-/**
- * The curated catalog. Keys are the vocabulary the model is taught; several are
- * semantic aliases for the same geometry (decision/diamond), because the model will
- * reach for both names.
- */
-export const SHAPE_CATALOG: Record<string, ShapeSpec> = {
-    // ---- the rectangle family (the original six) ----
-    box: { style: "rounded=0;", textScale: 1 },
-    round: { style: "rounded=1;arcSize=12;", textScale: 1 },
-    terminator: { style: "rounded=1;arcSize=50;", textScale: 1.15 },
-    decision: {
-        style: "rhombus;perimeter=rhombusPerimeter;",
-        textScale: 1.5,
-    },
-    diamond: {
-        style: "rhombus;perimeter=rhombusPerimeter;",
-        textScale: 1.5,
-    },
-    data: {
-        style: "shape=parallelogram;perimeter=parallelogramPerimeter;fixedSize=1;size=14;",
-        textScale: 1.2,
-    },
-    document: { style: "shape=document;boundedLbl=1;", textScale: 1.15 },
-
-    // ---- the semantic vocabulary (D2's tier: a node that IS a thing) ----
-    /** A database or datastore. */
-    cylinder: {
-        style: "shape=cylinder3;boundedLbl=1;backgroundOutline=1;size=12;",
-        textScale: 1.3,
-    },
-    /** A message queue: a cylinder on its side. */
-    queue: {
-        style: "shape=cylinder3;direction=south;boundedLbl=1;backgroundOutline=1;size=12;",
-        textScale: 1.3,
-    },
-    /** An actor or user. Label below the figure. */
-    person: {
-        style: "shape=umlActor;verticalLabelPosition=bottom;verticalAlign=top;outlineConnect=0;",
-        textScale: 1,
-        labelOutside: true,
-        glyph: { w: 40, h: 60 },
-    },
-    /** An external system, the internet. */
-    cloud: { style: "ellipse;shape=cloud;", textScale: 1.6 },
-    /** A service or process step. */
-    hexagon: {
-        style: "shape=hexagon;perimeter=hexagonPerimeter2;fixedSize=1;size=16;",
-        textScale: 1.25,
-    },
-    /** A concept, state or category. */
-    ellipse: { style: "ellipse;", textScale: 1.3 },
-    /** A speech-bubble annotation. */
-    callout: {
-        style: "shape=callout;perimeter=calloutPerimeter;rounded=1;size=16;position=0.5;base=24;",
-        textScale: 1.35,
-    },
-    /** A chevron stage in a pipeline. */
-    step: {
-        style: "shape=step;perimeter=stepPerimeter;fixedSize=1;size=16;",
-        textScale: 1.2,
-    },
-    /** A sticky note. */
-    note: { style: "shape=note;size=14;", textScale: 1.1 },
-    /** A card with a cut corner. */
-    card: { style: "shape=card;size=14;", textScale: 1.1 },
-    /** A process box with side bars (predefined subroutine). */
-    process: { style: "shape=process;size=0.1;", textScale: 1.2 },
-    /** Punched tape — legacy data, files. */
-    tape: { style: "shape=tape;size=0.2;", textScale: 1.3 },
-    /** A double-walled cube. */
-    cube: { style: "shape=cube;size=12;", textScale: 1.25 },
-}
-
-/** A shape token that may pass through unrecognised: draw.io style-key charset only. */
-const SAFE_TOKEN = /^[a-zA-Z0-9._]+$/
-
-export interface ResolvedShape {
-    spec: ShapeSpec
-    /** Set when the token was not in the catalog and passed through verbatim. */
-    passthrough?: boolean
-}
-
-/**
- * Resolve a shape token: catalog entry, safe pass-through, or null for a token that
- * could inject style keys (`;`/`=`/quotes) and must be rejected outright.
- */
-export function resolveShape(token: string): ResolvedShape | null {
-    const known = SHAPE_CATALOG[token]
-    if (known) return { spec: known }
-    if (!SAFE_TOKEN.test(token)) return null
-    // Unknown but safe: emit verbatim. draw.io degrades an unregistered shape to a
-    // rectangle, so the worst case is a plain box — same as before the vocabulary
-    // existed. The conservative scale covers roughly-convex real shapes.
-    return {
-        spec: { style: `shape=${token};`, textScale: 1.25 },
-        passthrough: true,
-    }
-}
-
-/** The catalog key most similar to a token, for "did you mean" hints. */
-export function nearestShape(token: string): string | null {
-    const t = token.toLowerCase()
-    let best: string | null = null
-    let bestD = 3 // more than 2 edits away is not a typo
-    for (const key of Object.keys(SHAPE_CATALOG)) {
-        const d = editDistance(t, key.toLowerCase(), bestD)
-        if (d < bestD) {
-            bestD = d
-            best = key
-        }
-    }
-    return best
-}
-
-/** Bounded Levenshtein distance; returns limit when the strings are further apart. */
-function editDistance(a: string, b: string, limit: number): number {
-    if (Math.abs(a.length - b.length) >= limit) return limit
-    const prev = new Array(b.length + 1)
-    for (let j = 0; j <= b.length; j++) prev[j] = j
-    for (let i = 1; i <= a.length; i++) {
-        let diag = prev[0]
-        prev[0] = i
-        let rowMin = prev[0]
-        for (let j = 1; j <= b.length; j++) {
-            const cur = Math.min(
-                prev[j] + 1,
-                prev[j - 1] + 1,
-                diag + (a[i - 1] === b[j - 1] ? 0 : 1),
-            )
-            diag = prev[j]
-            prev[j] = cur
-            if (cur < rowMin) rowMin = cur
-        }
-        if (rowMin >= limit) return limit
-    }
-    return Math.min(prev[b.length], limit)
-}
-
-// ---- structured style merge ----
-
-/**
- * Merge style fragments by key, later fragments winning.
- *
- * A draw.io style is `tok;key=value;key=value;` — bare class tokens (rhombus, ellipse,
- * text) come first and key=value pairs follow. String concatenation made every
- * conflict a duplicate key resolved by draw.io's last-wins rule, which worked until
- * shape fragments and theme fragments both owned geometry keys (a theme's rounded=1
- * against a shape's rhombus). Merging structurally keeps exactly one token per key and
- * one bare-token set, so the output is canonical and the ownership rule — theme owns
- * colour and type, shape owns geometry — is enforced by fragment ORDER, not by hoping
- * the keys never meet.
- *
- * Bare tokens are kept in first-appearance order, except that a later fragment's bare
- * SHAPE CLASS (rhombus/ellipse/triangle) replaces an earlier one — two shape classes
- * on one cell is a contradiction, not a union.
- */
-export function mergeStyle(...fragments: (string | undefined)[]): string {
-    const bare: string[] = []
-    const kv = new Map<string, string>()
-    const SHAPE_CLASSES = new Set(["rhombus", "ellipse", "triangle"])
-    for (const f of fragments) {
-        if (!f) continue
-        for (const tok of f.split(";")) {
-            if (tok === "") continue
-            const eq = tok.indexOf("=")
-            if (eq < 0) {
-                if (SHAPE_CLASSES.has(tok)) {
-                    const i = bare.findIndex((b) => SHAPE_CLASSES.has(b))
-                    if (i >= 0) bare.splice(i, 1)
-                }
-                if (!bare.includes(tok)) bare.push(tok)
-                continue
-            }
-            const key = tok.slice(0, eq)
-            kv.set(key, tok.slice(eq + 1))
-            // An explicit shape= also displaces a bare shape class from an earlier
-            // fragment — same contradiction as two bare classes.
-            if (key === "shape") {
-                const i = bare.findIndex((b) => SHAPE_CLASSES.has(b))
-                if (i >= 0) bare.splice(i, 1)
-            }
-        }
-    }
-    let out = bare.join(";")
-    if (out) out += ";"
-    for (const [k, v] of kv) out += `${k}=${v};`
-    return out
-}

lib/diagram-engine/theme.ts

@@ -1,265 +0,0 @@
-/**
- * The theme: design tokens plus one composition rule, in place of style tables.
- *
- * The engine's original deal was: the model declares structure, the engine computes
- * geometry. But every box rendered identically — white, 11px, black border — so anything
- * whose meaning lives in visual hierarchy (a paper-summary poster, a cheat sheet, a
- * comparison panel) came out flat, and the only escape was hand-written XML with no layout
- * guarantees at all.
- *
- * Two ideas fix that generally, not per diagram type:
- *
- *   ROLE — what a node IS in the information hierarchy: a masthead, a section heading, a
- *   key number, fine print. The model judges this well. Each role maps to a type size and
- *   an emphasis (filled / outlined / ghost), not to any colour.
- *
- *   GROUP — which semantic zone a node belongs to: remote vs local, one poster section vs
- *   another. Each distinct group name gets one HUE RAMP — a light tint, a mid stroke, a
- *   dark text colour — assigned in order of first appearance.
- *
- * `themedStyle(role, hue, kind)` composes the two by rule. A heading container in group 2
- * gets that hue's tint as its panel and the dark step for its title; a metric in the same
- * group gets the mid step as a heavy border. Nothing is enumerated per combination, so a
- * new diagram kind gets full theming by tagging its nodes — there is no table to extend.
- * The model never sees a hex value; the same declaration always renders the same way.
- */
-
-/** What a node is, in the information hierarchy of the diagram. */
-export type Role =
-    | "banner" //   the masthead: large type on the theme's one dark field
-    | "heading" //  a section title / titled panel
-    | "body" //     ordinary content (the default look)
-    | "callout" //  something the reader must not miss
-    | "good" //     a positive verdict (always green, group or not)
-    | "bad" //      a negative verdict or warning (always red)
-    | "metric" //   the key number
-    | "muted" //    fine print
-
-export const ROLES: readonly Role[] = [
-    "banner",
-    "heading",
-    "body",
-    "callout",
-    "good",
-    "bad",
-    "metric",
-    "muted",
-]
-
-export function isRole(v: string | null | undefined): v is Role {
-    return ROLES.includes(v as Role)
-}
-
-// ---- tokens ----
-
-/** One hue, three steps: a field to sit on, a line to draw with, a colour to write in. */
-export interface HueRamp {
-    tint: string
-    base: string
-    dark: string
-}
-
-/**
- * The hue ramps groups draw from, in assignment order.
- *
- * Tint/base pairs are draw.io's classic palette, so themed output looks native to the
- * editor; the dark step is the same hue pulled down far enough for 4.5:1 text on white.
- */
-export const HUES: readonly HueRamp[] = [
-    { tint: "#DAE8FC", base: "#6C8EBF", dark: "#1A237E" }, // blue
-    { tint: "#D5E8D4", base: "#82B366", dark: "#1B5E20" }, // green
-    { tint: "#FFE6CC", base: "#D79B00", dark: "#8A5A00" }, // orange
-    { tint: "#E1D5E7", base: "#9673A6", dark: "#4A2E5E" }, // purple
-    { tint: "#F8CECC", base: "#B85450", dark: "#7F1D1D" }, // red
-    { tint: "#FFF2CC", base: "#D6B656", dark: "#7A5C00" }, // yellow
-]
-
-/** The neutral ramp, for ungrouped nodes: today's grey-on-white look. */
-export const NEUTRAL: HueRamp = {
-    tint: "#F5F8FB",
-    base: "#5A6B7B",
-    dark: "#1A1A1A",
-}
-
-/** Semantic verdict hues: good is green and bad is red no matter what group says. */
-const GOOD: HueRamp = { tint: "#D5E8D4", base: "#82B366", dark: "#1B5E20" }
-const BAD: HueRamp = { tint: "#F8CECC", base: "#B85450", dark: "#7F1D1D" }
-/** The callout field: a warm highlight distinct from every group tint. */
-const CALLOUT: HueRamp = { tint: "#FFF9C4", base: "#B8860B", dark: "#6D4C00" }
-
-/** Type scale, px. One scale for every diagram kind. */
-export const TYPE = { xs: 9, sm: 11, md: 13, lg: 15, xl: 20 } as const
-
-/** The hue ramp for the n-th distinct group. Wraps: a 7th group reuses the 1st hue. */
-export function hueOf(index: number): HueRamp {
-    return HUES[index % HUES.length]
-}
-
-// ---- the composition rule ----
-
-/** How a role renders, independent of colour. */
-interface RoleSpec {
-    size: number
-    bold: boolean
-    /** filled: dark field, light text. tinted: hue field. outlined: white field, hue border.
-     *  ghost: no field, no border — pure text. */
-    emphasis: "filled" | "tinted" | "outlined" | "ghost"
-    /** Overrides the group hue; verdicts stay green/red whatever zone they sit in. */
-    hue?: HueRamp
-    /** Fill the container's cross axis, the way a masthead spans its page. */
-    stretch?: boolean
-    /** Minimum cell height. */
-    minH: number
-    /** Character width relative to 11px type, for the measure pass. */
-    charScale: number
-}
-
-/** The masthead field when no group says otherwise: the deep navy of the first hue. */
-const BANNER: HueRamp = { tint: "#DAE8FC", base: "#6C8EBF", dark: "#1A237E" }
-
-const ROLE_SPECS: Record<Role, RoleSpec> = {
-    banner: {
-        size: TYPE.xl,
-        bold: true,
-        emphasis: "filled",
-        hue: BANNER,
-        stretch: true,
-        minH: 64,
-        charScale: 1.8,
-    },
-    heading: {
-        size: TYPE.lg,
-        bold: true,
-        emphasis: "ghost",
-        stretch: true,
-        minH: 32,
-        charScale: 1.35,
-    },
-    body: {
-        size: TYPE.sm,
-        bold: false,
-        emphasis: "outlined",
-        minH: 44,
-        charScale: 1,
-    },
-    callout: {
-        size: TYPE.sm,
-        bold: true,
-        emphasis: "tinted",
-        hue: CALLOUT,
-        minH: 44,
-        charScale: 1,
-    },
-    good: {
-        size: TYPE.sm,
-        bold: false,
-        emphasis: "tinted",
-        hue: GOOD,
-        minH: 44,
-        charScale: 1,
-    },
-    bad: {
-        size: TYPE.sm,
-        bold: false,
-        emphasis: "tinted",
-        hue: BAD,
-        minH: 44,
-        charScale: 1,
-    },
-    metric: {
-        size: TYPE.xl,
-        bold: true,
-        emphasis: "outlined",
-        minH: 56,
-        charScale: 1.8,
-    },
-    muted: {
-        size: TYPE.xs,
-        bold: false,
-        emphasis: "ghost",
-        minH: 24,
-        charScale: 0.82,
-    },
-}
-
-/**
- * Does this role already draw itself with no border?
- *
- * The parser needs this to tell a THEME's `strokeColor=none` from a DECLARED one. A banner is
- * a dark filled slab and a heading is ghost text; both are borderless because of what they
- * are, not because anyone asked. Recording that as an explicit override would make it
- * outlive a later role change, since `set_role` clears a node's style but keeps its text
- * overrides.
- *
- * Leaf only: the container branch of `themedStyle` always draws a border, whatever the role.
- */
-export function roleIsBorderless(
-    role: Role | undefined,
-    kind: "leaf" | "container",
-): boolean {
-    if (kind === "container") return false
-    const e = ROLE_SPECS[role ?? "body"].emphasis
-    return e === "filled" || e === "ghost"
-}
-
-/** Metrics the measure pass needs, so layout reserves what render will draw. */
-export function roleMetrics(role: Role | undefined): {
-    fontSize: number
-    minH: number
-    charScale: number
-    stretch: boolean
-} {
-    const s = ROLE_SPECS[role ?? "body"]
-    return {
-        fontSize: s.size,
-        minH: s.minH,
-        charScale: s.charScale,
-        stretch: s.stretch === true,
-    }
-}
-
-/**
- * The style tokens for one node: the whole theme in a single rule.
- *
- * `hue` is the node's group ramp (or the neutral ramp); a role with a semantic hue
- * (good/bad/callout) overrides it. `kind` softens the treatment for containers — a
- * section panel is a field its children sit on, so it takes the tint at panel weight
- * rather than a leaf's full treatment.
- */
-export function themedStyle(
-    role: Role,
-    hue: HueRamp,
-    kind: "leaf" | "container",
-): string {
-    const spec = ROLE_SPECS[role]
-    const ramp = spec.hue ?? hue
-    const size = kind === "container" && role === "banner" ? TYPE.lg : spec.size
-    const font = `fontSize=${size};${spec.bold || kind === "container" ? "fontStyle=1;" : ""}`
-
-    if (spec.emphasis === "filled")
-        return `fillColor=${ramp.dark};strokeColor=none;fontColor=#FFFFFF;${font}rounded=1;arcSize=6;`
-
-    if (kind === "container") {
-        // A panel: the tint as a quiet field, the dark step for its title, the base for
-        // its border. This is where "every section gets its own colour" comes from —
-        // a heading container plus a group, no extra mechanism.
-        return `fillColor=${ramp.tint};strokeColor=${ramp.base};fontColor=${ramp.dark};${font}verticalAlign=top;align=left;spacingLeft=10;spacingTop=6;`
-    }
-
-    if (spec.emphasis === "ghost")
-        return `fillColor=none;strokeColor=none;fontColor=${ramp.dark};${font}align=left;`
-
-    if (spec.emphasis === "tinted") {
-        // A callout keeps a heavy left bar, the editor's convention for "note well".
-        const bar = role === "callout" ? `strokeWidth=2;` : ""
-        return `fillColor=${ramp.tint};strokeColor=${ramp.base};fontColor=${ramp.dark};${bar}${font}`
-    }
-
-    // outlined: the hue carried by the border and text. A grouped ordinary node takes its
-    // zone's tint as the field — colour-as-grouping is the whole point of naming a zone —
-    // while an ungrouped one stays white. A metric stays white either way, so its number
-    // sits on the page's calmest field with the hue in a heavy border.
-    const weight = role === "metric" ? "strokeWidth=2;" : ""
-    const field = role === "body" && ramp !== NEUTRAL ? ramp.tint : "#FFFFFF"
-    return `fillColor=${field};strokeColor=${ramp.base};fontColor=${ramp.dark};${weight}${font}`
-}

lib/diagram-engine/tw.ts

@@ -1,538 +0,0 @@
-/**
- * Tailwind utility classes → the engine's layout fields.
- *
- * WHY a second way to say the same thing. The engine's own vocabulary (`dir`, `grow`,
- * `align`, `justify`, `pad`, `gap`, `maxW`) is words we invented, so a model has seen them
- * only in our tool description. It has seen `flex-col grow-3 items-stretch p-4` millions of
- * times. Microsoft's DSL study (arXiv 2407.02742) found models hallucinate custom function
- * names at a much higher rate than familiar ones, and arXiv 2311.09519 measured a large
- * improvement from swapping a rare DSL for a popular language, precisely because it puts the
- * output back in the distribution the model was trained on.
- *
- * The other half of why Tailwind and not free-form CSS: its values are a FIXED SCALE, not
- * arbitrary numbers. `p-4` is 16px because one spacing unit is 4px, and there is no `p-7.5`.
- * Tailwind's own docs make that the point of the thing — with inline styles "every value is a
- * magic number", with utilities you pick from a system. That is the property we want, because
- * an unconstrained number field is exactly where a model invents 13px here and 27px there.
- *
- * The supported set was picked by reading Tailwind's property index against the draw.io
- * renderer's ACTUAL SOURCE — public/drawio/mxgraph/src and public/drawio/js/grapheditor,
- * vendored in this repo — rather than against a prose style reference. That matters: three
- * properties were excluded on wrong grounds when the reference was a document, and reading
- * the code put them back (radius, strikethrough, shadow, all noted below).
- *
- * WHAT IS DELIBERATELY NOT HERE, and why:
- *
- *   - COLOUR of any kind (`bg-*`, `text-red-500`, `border-blue-400`). draw.io has
- *     `fillColor`/`fontColor`/`strokeColor`, so this is possible — but colour is derived from
- *     `role` and `group` precisely so one palette stays coherent, and a colour class would be
- *     a back door into the hex-picking that was removed. Gradients (`bg-linear-to-b from-X
- *     to-Y`) are excluded for the same reason, even though `gradientColor` with a four-way
- *     `gradientDirection` maps onto them exactly (mxShape.js:1392-1393, 1054-1060).
- *
- *   - Per-SIDE borders (`border-t`, `border-l-4`, `border-x`). draw.io draws these properly:
- *     `shape=partialRectangle` reads independent `top`/`right`/`bottom`/`left` booleans
- *     (Shapes.js:3914-3917) and still fills the background first (3919-3920), so a single
- *     heavy left edge would render correctly. The cost is the SHAPE SLOT: `partialRectangle`
- *     is itself a shape name, so a node could not be both a diamond and left-edge-only. What
- *     a node IS — a database, a decision, a person — outranks how its border looks, so the
- *     shape vocabulary keeps the slot.
- *
- *   - Per-SIDE padding (`pt-8`, `px-4`). draw.io's `spacingTop`/`spacingRight`/
- *     `spacingBottom`/`spacingLeft` (mxText.js:422-425) look like an exact match and are not:
- *     they pad the LABEL inside its own cell, while this engine's `pad` is the room a
- *     container leaves for its CHILDREN. Accepting `pt-8` would suggest it pushes child nodes
- *     down, which it cannot.
- *
- *   - `outline-*` (width, colour, style, offset). draw.io has no concept: a shape carries one
- *     border, and nothing draws a second ring outside it. In CSS an outline is a focus ring,
- *     which a static diagram does not have.
- *
- *   - `opacity-*`. draw.io's `opacity` is 0–100 and would map cleanly, but Tailwind's
- *     `opacity-<number>` takes ANY number — `opacity-37` is valid — so it is not a scale.
- *     Admitting it would give up the one property that makes this vocabulary worth having.
- *
- *   - `truncate` / `text-ellipsis`. Sets `text-overflow: ellipsis`. draw.io's `overflow`
- *     branches on exactly five values — visible, hidden, fill, width, block (mxText.js:
- *     1080-1095) — and a repo-wide grep for "ellipsis" finds no implementation, so the text
- *     would be cut with no "…": a class named `truncate` that silently loses characters.
- *
- *   - Seven of the nine `font-*` weights. See UNSUPPORTED_WEIGHTS below.
- *
- *   - `text-shadow-*`. Unlike the box `shadow-*` family, draw.io's `textShadow`
- *     (mxText.js:668) is a bare on/off flag with no offset or blur, so Tailwind's six sizes
- *     would collapse into one picture.
- *
- *   - Per-CORNER radius (`rounded-tl-lg`) and the decorative corner treatments beside it
- *     (snip, fold, inverse round). draw.io does have these, but only on a separate template
- *     shape, `mxgraph.basic.rect` (Shapes.js:4118), which would take the place of the node's
- *     own `shape` — the same trade the per-side borders lose. Whole-shape `rounded-*` IS
- *     supported and costs no slot; see RADIUS.
- *
- *   - `tracking-*` (letter-spacing), `uppercase`/`lowercase`/`capitalize` (text-transform),
- *     and per-node `leading-*` (line-height). Not merely coarse — absent. Grepping the whole
- *     vendored renderer for letterSpacing/textTransform finds nothing, and line height is a
- *     global constant (`mxConstants.LINE_HEIGHT`) with no per-cell style key.
- *
- *   - `rotate-*`, `scale-*`, `skew-*`, `translate-*`. draw.io has `rotation`/`flipH`/`flipV`,
- *     but a rotated box breaks the two things this engine guarantees: the layout no longer
- *     knows what area it covers, and the edge router cannot route around it.
- *
- *   - Document-flow properties (`float`, `clear`, `position`, `top/right/bottom/left`,
- *     `z-index`, `visibility`, `columns`, `break-*`, `object-*`, `overscroll-*`) and the
- *     table and list families. There is no document flow here — every coordinate is computed
- *     — and draw.io has no z-index at all: later cells simply paint on top.
- *
- *   - `filter`/`backdrop-filter`, `mask-*`, `mix-blend-mode`, `transition-*`, `animation`,
- *     `perspective*`, `cursor`, `resize`, `appearance`, `caret-color`, `accent-color`:
- *     no corresponding key anywhere in the vendored renderer.
- *
- *   - Arbitrary values (`w-[137px]`, `p-[13px]`). The scale is the feature; a bracket escape
- *     hatch removes it.
- *
- * Unknown classes are returned in `ignored` rather than rejected — D2's "warnings over
- * errors" rule: a diagram that renders with one class dropped beats an error that renders
- * nothing. The caller reports them, which is how a typo becomes a one-turn fix instead of a
- * silent no-op.
- */
-
-import type { Align, Justify } from "./types"
-
-/** What a class string resolves to. Every field optional: a class string sets only what it names. */
-export interface TwLayout {
-    dir?: "row" | "col"
-    grow?: number
-    align?: Align
-    justify?: Justify
-    alignItems?: Align
-    gap?: number
-    pad?: number
-    maxW?: number
-    /** `min-w-0`: let a weight shrink this below its content width. */
-    minW0?: boolean
-
-    // ---- text, the part draw.io can actually render ----
-    /** `font-bold` / `font-normal`. draw.io has one bold bit, not nine weights. */
-    bold?: boolean
-    /** `italic` / `not-italic`. */
-    italic?: boolean
-    /** `underline` / `no-underline`. */
-    underline?: boolean
-    /** `line-through`. draw.io's fontStyle carries a strikethrough bit beside the other three. */
-    strike?: boolean
-    /** `text-xs`…`text-4xl` → px, from Tailwind's own scale. */
-    fontSize?: number
-    /** `text-left` / `text-center` / `text-right`. */
-    textAlign?: "left" | "center" | "right"
-    /** `align-top` / `align-middle` / `align-bottom`. */
-    verticalAlign?: "top" | "middle" | "bottom"
-    /** `whitespace-nowrap` / `whitespace-normal`. */
-    nowrap?: boolean
-
-    // ---- border ----
-    /** `border` / `border-N` → strokeWidth in px. */
-    borderWidth?: number
-    /** `border-dashed` / `border-dotted` / `border-solid`. */
-    borderStyle?: "solid" | "dashed" | "dotted"
-    /** `rounded`, `rounded-lg`, `rounded-full` → corner radius in px. */
-    radius?: number
-    /** `border-none` / `border-0`. */
-    borderless?: boolean
-    /** `shadow-sm`…`shadow-xl` → 1–4; `shadow-none` → 0. See SHADOW. */
-    shadow?: number
-
-    /** Classes that matched nothing, verbatim and in order. */
-    ignored: string[]
-}
-
-/**
- * Tailwind's spacing scale: one unit is 0.25rem, which is 4px at the default root size.
- *
- * Only whole steps are accepted. Tailwind itself has half-steps (`p-0.5`), but a diagram has
- * no use for 2px padding and allowing them widens the scale for nothing.
- */
-const SPACING_UNIT = 4
-
-/** `p-6` / `gap-3` → px, or null when the suffix is not a plain scale step. */
-function scaleToPx(suffix: string): number | null {
-    if (!/^\d+$/.test(suffix)) return null
-    return Number(suffix) * SPACING_UNIT
-}
-
-/**
- * Tailwind's width fractions, as a share of the parent.
- *
- * Expressed as `grow` rather than an absolute width, because that is what the fraction means
- * inside a flex row: `w-1/3` beside `w-2/3` is the same layout as `grow-1` beside `grow-2`,
- * and going through grow means the existing proportional path applies — including the rule
- * that a declared cap outranks it.
- */
-function fractionToGrow(suffix: string): number | null {
-    const m = /^(\d+)\/(\d+)$/.exec(suffix)
-    if (!m) return null
-    const num = Number(m[1])
-    const den = Number(m[2])
-    if (den === 0 || num === 0 || num > den) return null
-    return num
-}
-
-const ALIGN_ITEMS: Record<string, Align> = {
-    "items-start": "start",
-    "items-center": "center",
-    "items-end": "end",
-    "items-stretch": "stretch",
-}
-
-const ALIGN_SELF: Record<string, Align> = {
-    "self-start": "start",
-    "self-center": "center",
-    "self-end": "end",
-    "self-stretch": "stretch",
-}
-
-const JUSTIFY: Record<string, Justify> = {
-    "justify-start": "start",
-    "justify-center": "center",
-    "justify-end": "end",
-    "justify-between": "between",
-    "justify-around": "around",
-    "justify-evenly": "evenly",
-}
-
-/**
- * Tailwind's type scale in px, its own documented values.
- *
- * Stops at 4xl. The ladder goes on to 9xl (128px), but a 128px word is not a diagram
- * label, and offering the step invites a model to pick it.
- */
-const FONT_SIZE: Record<string, number> = {
-    "text-xs": 12,
-    "text-sm": 14,
-    "text-base": 16,
-    "text-lg": 18,
-    "text-xl": 20,
-    "text-2xl": 24,
-    "text-3xl": 30,
-    "text-4xl": 36,
-}
-
-/**
- * `text-left|center|right` — horizontal text alignment inside the shape.
- *
- * `text-justify`, `text-start` and `text-end` are absent because draw.io's `align` has
- * only the three physical values; justified text is not available at all.
- */
-const TEXT_ALIGN: Record<string, "left" | "center" | "right"> = {
-    "text-left": "left",
-    "text-center": "center",
-    "text-right": "right",
-}
-
-/** `align-*` → draw.io's verticalAlign. */
-const VERTICAL_ALIGN: Record<string, "top" | "middle" | "bottom"> = {
-    "align-top": "top",
-    "align-middle": "middle",
-    "align-bottom": "bottom",
-}
-
-/**
- * Tailwind's border-radius scale in px, its own documented values.
- *
- * These are REAL pixels, which is only true because of `absoluteArcSize`: draw.io's `arcSize`
- * is a percentage of the shape by default, but that flag switches it to absolute units
- * (mxShape.js:1172-1189). Without it a radius class would mean something different on every
- * box, which is why this looked unimplementable at first glance.
- *
- * `rounded-full` is `calc(infinity * 1px)` in Tailwind v4 — "as round as it goes". The same
- * function clamps the radius to half the shorter side, so any number past half the box's
- * height gives a stadium. 200 is chosen rather than something enormous because the number
- * reaches the user: draw.io's Arrange panel shows `arcSize` in an editable field, and a
- * diagram box taller than 400px does not exist, so 200 is both always enough and readable.
- */
-const RADIUS: Record<string, number> = {
-    "rounded-none": 0,
-    "rounded-xs": 2,
-    "rounded-sm": 4,
-    rounded: 4,
-    "rounded-md": 6,
-    "rounded-lg": 8,
-    "rounded-xl": 12,
-    "rounded-2xl": 16,
-    "rounded-3xl": 24,
-    "rounded-4xl": 32,
-    "rounded-full": 200,
-}
-
-/**
- * Tailwind's box-shadow steps, as a rung number the renderer turns into draw.io's five
- * shadow parameters. 0 means "explicitly no shadow".
- *
- * draw.io's shadow is not the on/off flag it looks like: `shadowOffsetX`, `shadowOffsetY`,
- * `shadowBlur`, `shadowColor` and `shadowOpacity` are read independently
- * (mxShape.js:505-535) and become a CSS `drop-shadow(dx dy blur colour)` (540-552). Since
- * Tailwind's own steps are also just offset-and-blur, they map one for one.
- *
- * Four rungs, not Tailwind's eight. `shadow-2xs` and `shadow-xs` are indistinguishable from
- * `shadow-sm` at a diagram's scale, and `shadow-2xl`'s 50px blur is noise on a page of
- * boxes — offering a step invites a model to pick it.
- */
-const SHADOW: Record<string, number> = {
-    "shadow-none": 0,
-    "shadow-sm": 1,
-    "shadow-md": 2,
-    "shadow-lg": 3,
-    "shadow-xl": 4,
-}
-
-/**
- * Font-weight classes that are NOT accepted, and why.
- *
- * Tailwind has nine weights; draw.io's `fontStyle` is a bitmask whose bold flag is a single
- * bit. Accepting all nine would collapse five of them onto "bold" and four onto "normal",
- * which is the same defect that rules out `shadow-*` (six sizes, one on/off flag). So only
- * `font-bold` and `font-normal` are honoured and the rest are reported, rather than
- * pretending a distinction the renderer cannot draw.
- */
-const UNSUPPORTED_WEIGHTS = new Set([
-    "font-thin",
-    "font-extralight",
-    "font-light",
-    "font-medium",
-    "font-semibold",
-    "font-extrabold",
-    "font-black",
-])
-
-/**
- * Parse a Tailwind class string into layout fields.
- *
- * Later classes win over earlier ones, the same as Tailwind's own last-one-wins behaviour
- * for conflicting utilities, so a caller can append an override without removing anything.
- */
-export function parseTw(classes: string): TwLayout {
-    const out: TwLayout = { ignored: [] }
-    for (const raw of String(classes ?? "").split(/\s+/)) {
-        const cls = raw.trim()
-        if (!cls) continue
-
-        // Direction. `flex` on its own is the default and says nothing here — every engine
-        // container is already a flex container — so it is accepted and ignored rather than
-        // reported, since a model writing `flex flex-col` is not making a mistake.
-        if (cls === "flex" || cls === "flex-row") {
-            if (cls === "flex-row") out.dir = "row"
-            continue
-        }
-        if (cls === "flex-col") {
-            out.dir = "col"
-            continue
-        }
-
-        if (cls in ALIGN_ITEMS) {
-            out.alignItems = ALIGN_ITEMS[cls]
-            continue
-        }
-        if (cls in ALIGN_SELF) {
-            out.align = ALIGN_SELF[cls]
-            continue
-        }
-        if (cls in JUSTIFY) {
-            out.justify = JUSTIFY[cls]
-            continue
-        }
-
-        // `grow` alone is flex-grow: 1, `grow-N` is the weight. Tailwind writes the latter
-        // as `grow-[3]`; the plain form is accepted because it is what a model reaches for
-        // and the bracket form carries no extra meaning here.
-        if (cls === "grow") {
-            out.grow = 1
-            continue
-        }
-        const growN = /^grow-(\d+)$/.exec(cls)
-        if (growN) {
-            out.grow = Number(growN[1])
-            continue
-        }
-        // `flex-1` / `flex-3`: the shorthand whose whole point is proportional sizing.
-        const flexN = /^flex-(\d+)$/.exec(cls)
-        if (flexN) {
-            out.grow = Number(flexN[1])
-            continue
-        }
-
-        // Fractional widths become weights — see fractionToGrow.
-        const wFrac = /^w-(\d+\/\d+)$/.exec(cls)
-        if (wFrac) {
-            const g = fractionToGrow(wFrac[1])
-            if (g !== null) {
-                out.grow = g
-                continue
-            }
-        }
-        if (cls === "w-full") {
-            out.align = "stretch"
-            continue
-        }
-
-        // `min-w-0` is the standard CSS escape hatch for "let the weight win over my
-        // content width". Without it a weighted child is floored by its own text — that is
-        // real flexbox behaviour, since `min-width` defaults to `auto` — so a narrow column
-        // beside a wide one settles at its text width and a declared 2:1 comes out 1.4:1.
-        if (cls === "min-w-0") {
-            out.minW0 = true
-            continue
-        }
-
-        // Spacing. `p-*` is interior padding, `gap-*` the space between children. Tailwind's
-        // per-side variants (`pt-*`, `px-*`) are not here: the engine has one padding value,
-        // and quietly treating `pt-8` as padding on all four sides would be wrong in a way
-        // the model could not see.
-        const pad = /^p-(\d+)$/.exec(cls)
-        if (pad) {
-            const px = scaleToPx(pad[1])
-            if (px !== null) {
-                out.pad = px
-                continue
-            }
-        }
-        const gap = /^gap-(\d+)$/.exec(cls)
-        if (gap) {
-            const px = scaleToPx(gap[1])
-            if (px !== null) {
-                out.gap = px
-                continue
-            }
-        }
-
-        // `max-w-*` uses the spacing scale too, so `max-w-96` is 384px. Tailwind's named
-        // sizes are also accepted, because a model reaches for `max-w-md` more readily than
-        // for a step number.
-        const maxW = /^max-w-(\d+)$/.exec(cls)
-        if (maxW) {
-            const px = scaleToPx(maxW[1])
-            if (px !== null) {
-                out.maxW = px
-                continue
-            }
-        }
-        const named = NAMED_MAX_W[cls]
-        if (named) {
-            out.maxW = named
-            continue
-        }
-
-        // ---- text ----
-        // The three flags draw.io's fontStyle bitmask actually carries. They combine by
-        // adding bits, so bold + italic is legal and needs no special case here.
-        if (cls === "font-bold" || cls === "font-normal") {
-            out.bold = cls === "font-bold"
-            continue
-        }
-        // The other seven weights fall through to `ignored` on purpose, so the model is
-        // told the distinction was dropped instead of quietly getting plain bold.
-        if (UNSUPPORTED_WEIGHTS.has(cls)) {
-            out.ignored.push(cls)
-            continue
-        }
-        if (cls === "italic" || cls === "not-italic") {
-            out.italic = cls === "italic"
-            continue
-        }
-        if (cls === "underline" || cls === "no-underline") {
-            out.underline = cls === "underline"
-            continue
-        }
-        // Strikethrough is its own bit (8) beside bold/italic/underline, so it combines with
-        // them rather than replacing one. `no-underline` above deliberately does NOT clear
-        // it: in CSS both are values of `text-decoration-line`, and Tailwind's `no-underline`
-        // means "not underlined", not "undecorated".
-        if (cls === "line-through") {
-            out.strike = true
-            continue
-        }
-        // `text-*` is three different Tailwind properties sharing one prefix: size
-        // (text-lg), alignment (text-left) and COLOUR (text-red-500). The size and
-        // alignment tables are exact-match, so a colour class falls through to `ignored`
-        // rather than being mistaken for a size.
-        if (cls in FONT_SIZE) {
-            out.fontSize = FONT_SIZE[cls]
-            continue
-        }
-        if (cls in TEXT_ALIGN) {
-            out.textAlign = TEXT_ALIGN[cls]
-            continue
-        }
-        if (cls in VERTICAL_ALIGN) {
-            out.verticalAlign = VERTICAL_ALIGN[cls]
-            continue
-        }
-        if (cls === "whitespace-nowrap" || cls === "whitespace-normal") {
-            out.nowrap = cls === "whitespace-nowrap"
-            continue
-        }
-
-        // ---- border ----
-        // `border` alone is 1px, `border-N` is N px — Tailwind's border width is a plain
-        // pixel count, not the 4px spacing scale.
-        if (cls === "border") {
-            out.borderWidth = 1
-            continue
-        }
-        // `border-0` and `border-none` both mean no border, so they are handled before the
-        // numeric case (which would otherwise read border-0 as a zero-width border and
-        // leave draw.io drawing its default hairline).
-        if (cls === "border-none" || cls === "border-0") {
-            out.borderless = true
-            continue
-        }
-        const bw = /^border-(\d+)$/.exec(cls)
-        if (bw) {
-            out.borderWidth = Number(bw[1])
-            continue
-        }
-        if (
-            cls === "border-solid" ||
-            cls === "border-dashed" ||
-            cls === "border-dotted"
-        ) {
-            out.borderStyle = cls.slice("border-".length) as
-                | "solid"
-                | "dashed"
-                | "dotted"
-            continue
-        }
-        // Whole-shape corner radius. Per-corner classes (`rounded-tl-lg`) fall through to
-        // `ignored`: draw.io only offers those on a separate template shape.
-        if (cls in RADIUS) {
-            out.radius = RADIUS[cls]
-            continue
-        }
-
-        // Drop shadow. Per-side border classes (`border-l-4`) fall through to `ignored`, and
-        // so does every colour form (`shadow-blue-500`) since these tables are exact-match.
-        if (cls in SHADOW) {
-            out.shadow = SHADOW[cls]
-            continue
-        }
-
-        out.ignored.push(cls)
-    }
-    return out
-}
-
-/**
- * Tailwind's named max-width steps, in px.
- *
- * Tailwind's own values, rounded to whole pixels. Stopping at `4xl` is deliberate: anything
- * wider than about a thousand pixels is not a cap a diagram needs, and offering the whole
- * ladder invites a model to pick one at random.
- */
-const NAMED_MAX_W: Record<string, number> = {
-    "max-w-xs": 320,
-    "max-w-sm": 384,
-    "max-w-md": 448,
-    "max-w-lg": 512,
-    "max-w-xl": 576,
-    "max-w-2xl": 672,
-    "max-w-3xl": 768,
-    "max-w-4xl": 896,
-}

lib/diagram-engine/types.ts

@@ -1,444 +0,0 @@
-/**
- * The declarative node tree the layout engine works on.
- *
- * The model never writes coordinates. It declares nesting and direction; the engine
- * computes every x/y/width/height. The tree is not persisted anywhere — it is
- * re-derived from the canvas XML whenever it is needed (see parse.ts), so the canvas
- * stays the single source of truth and a user's manual edits are an input, never
- * something to be reconciled against a second copy of the state.
- */
-
-import type { Direction } from "./markers"
-import type { Role } from "./theme"
-
-export type { Direction } from "./markers"
-export type { Role } from "./theme"
-
-/**
- * Which cell of a swimlane pool a node sits in.
- *
- * `lane` indexes the role band, `col` the position along the flow. Cells are sparse:
- * nothing has to fill lane 1 column 3 for lane 2 column 3 to exist.
- */
-export interface PoolCell {
-    lane: number
-    col: number
-}
-
-/**
- * Cross-axis behaviour of a child inside a row/col group, CSS's align-items per child:
- * pin to either edge, centre (the default), or stretch to fill the axis.
- */
-export type Align = "start" | "center" | "end" | "stretch"
-
-/**
- * Presentation a node may override, beyond what its `role` decides.
- *
- * The admission test is that draw.io can draw the distinction FAITHFULLY — see tw.ts for the
- * properties that failed it and why. Most fields here are one style key with one value; a few
- * (`shadow`, `borderStyle`, the radius trio) expand to a fixed group of keys, which is fine
- * because the field still names one visual decision. What is not allowed is a field whose
- * values collapse onto fewer pictures than it promises.
- *
- * Kept as one optional object rather than a dozen loose fields so the round-trip has one
- * thing to carry and the node type does not grow a field per CSS property.
- *
- * `role` remains the primary way to say what a node IS; this is for the cases where the
- * model needs to override one aspect of how it looks.
- */
-export interface TextStyle {
-    /** Bold. draw.io's fontStyle carries one bold bit, not a weight ladder. */
-    bold?: boolean
-    italic?: boolean
-    underline?: boolean
-    /** Strikethrough — a fourth bit in the same mask, so it combines with the others. */
-    strike?: boolean
-    /** Type size in px. */
-    size?: number
-    /** Horizontal text alignment inside the shape. */
-    align?: "left" | "center" | "right"
-    /** Vertical text alignment inside the shape. */
-    valign?: "top" | "middle" | "bottom"
-    /** Keep the label on one line instead of wrapping it. */
-    nowrap?: boolean
-    /** Border thickness in px. */
-    borderWidth?: number
-    /** Border line style. Dashed and dotted read as "planned", "optional", "logical". */
-    borderStyle?: "solid" | "dashed" | "dotted"
-    /**
-     * Corner radius in px.
-     *
-     * Real pixels, not a percentage: draw.io's `arcSize` is a percentage of the shape by
-     * default, but `absoluteArcSize=1` switches it to absolute units, and it halves the
-     * value, so an 8px radius is emitted as `arcSize=16` (mxShape.getArcSize,
-     * mxShape.js:1172-1189).
-     *
-     * Overrides the radius of a shape that has one of its own: `round` and `terminator` are
-     * rounded rectangles already, and changing how round they are does not change what they
-     * are, so a radius class is allowed to win.
-     */
-    radius?: number
-    /** No border at all — a plain colour field. */
-    borderless?: boolean
-    /**
-     * Drop shadow, as a rung: 1–4 for Tailwind's sm/md/lg/xl, 0 for explicitly none.
-     *
-     * A rung rather than raw offsets because draw.io takes five separate numbers
-     * (`shadowOffsetX/Y`, `shadowBlur`, `shadowColor`, `shadowOpacity` — mxShape.js:505-535)
-     * and letting a caller set them individually is exactly the magic-number freedom this
-     * vocabulary exists to remove.
-     */
-    shadow?: number
-}
-
-/**
- * How a container spreads its children along its own stacking axis — CSS's
- * justify-content, and Yoga's six values.
- *
- * Until this existed the policy was hard-coded and differed per axis: a row padded its
- * gaps and centred the result, a column packed to the top and left every spare pixel in
- * one slab at the bottom. That slab is the empty bottom-left corner of a poster, and
- * nothing the model could declare would move it.
- */
-export type Justify =
-    | "start"
-    | "center"
-    | "end"
-    | "between"
-    | "around"
-    | "evenly"
-
-/**
- * What a box IS, drawn as its conventional outline.
- *
- * Open vocabulary: catalog names ("cylinder", "decision", "person"…) get full engine
- * support — correct perimeter, text sized to fit the outline. Any other draw.io shape
- * token passes through verbatim and degrades to a rectangle if the editor does not
- * know it. See shapes.ts.
- */
-export type BoxShape = string
-
-/** A catalog icon: a real stencil, drawn at a fixed glyph size with a label below. */
-export interface IconNode {
-    kind: "icon"
-    id: string
-    /** Catalog name, e.g. "s3" or "azure_virtual_machine". Resolved to a style by the catalog. */
-    name: string
-    label: string
-    /** Glyph size in px. Defaults to the diagram's icon size. */
-    size?: number
-    /** Verbatim style, when recovered from XML. Preferred over re-resolving `name`. */
-    style?: string
-    /** User froze this node's position — the engine must not move it. */
-    pinned?: boolean
-    /** Absolute geometry, when recovered from XML. Only meaningful for a pinned node. */
-    rect?: Rect
-    /** Position within a `pool` parent. Ignored elsewhere. */
-    cell?: PoolCell
-}
-
-/** A plain labelled rectangle, for things the catalog has no icon for. */
-export interface BoxNode {
-    kind: "box"
-    id: string
-    label: string
-    w?: number
-    h?: number
-    fill?: string
-    stroke?: string
-    bold?: boolean
-    /** What this node IS in the information hierarchy; the theme decides how that looks. */
-    role?: Role
-    /** Semantic zone name; every node sharing a group gets the same hue ramp. */
-    group?: string
-    /** Share of the parent's leftover flow-axis space, like flex-grow. 0/absent = natural size. */
-    grow?: number
-    /** Cross-axis behaviour within the parent. Absent = center; stretch = fill it. */
-    align?: Align
-    /**
-     * Hard cap on width, px. Text rewraps to fit instead of running the box wider, so
-     * this is what stops one long sentence stretching a whole page into a letterbox.
-     * Higher priority than `grow`, matching Yoga's min/max rule.
-     */
-    maxW?: number
-    /** Let a `grow` weight shrink this below its own text width — CSS's `min-width: 0`. */
-    minW0?: boolean
-    /** Presentation overrides: type, alignment, border. Absent means the role decides. */
-    text?: TextStyle
-    /** Flowchart outline. Absent means a plain rectangle. */
-    shape?: BoxShape
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-    /** Position within a `pool` parent. Ignored elsewhere. */
-    cell?: PoolCell
-}
-
-/** A page title. At most one per diagram; laid out outside the tree flow. */
-export interface TitleNode {
-    kind: "title"
-    id: string
-    label: string
-}
-
-/**
- * A container that stacks its children in one direction.
- *
- * `gname` is the catalog group stencil (group_vpc, group_region, …). When null the
- * container renders as a plain frame — a labelled rectangle with a border.
- */
-export interface GroupNode {
-    kind: "group"
-    id: string
-    gname: string | null
-    label: string
-    dir: Extract<Direction, "row" | "col">
-    gap: number
-    children: DiagramNode[]
-    fill?: string
-    stroke?: string
-    /** Section role; a themed panel for its children. */
-    role?: Role
-    /** Semantic zone name; the panel takes this hue's tint. */
-    group?: string
-    /** Share of the parent's leftover flow-axis space, like flex-grow. */
-    grow?: number
-    /** Cross-axis behaviour within the parent. Absent = center; stretch = fill it. */
-    align?: Align
-    /** How the children spread along `dir`. Absent = start (packed, no extra spacing). */
-    justify?: Justify
-    /** Cross-axis default for every child that does not declare its own `align`. */
-    alignItems?: Align
-    /** Hard cap on width, px. Children wrap or shrink to fit rather than overflow it. */
-    maxW?: number
-    /**
-     * Let a `grow` weight shrink this below its own content width — CSS's `min-width: 0`.
-     *
-     * Without it a weighted child is floored by its text, which is real flexbox behaviour
-     * (`min-width` defaults to `auto`) but means a declared 2:1 quietly resolves to
-     * whatever the two columns' text allows.
-     */
-    minW0?: boolean
-    /** Presentation overrides: title type, alignment, frame border. */
-    text?: TextStyle
-    /** Interior padding, px. Absent = the default (24). */
-    pad?: number
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-}
-
-/** A container that packs its children into a fixed number of columns. */
-export interface GridNode {
-    kind: "grid"
-    id: string
-    gname: string | null
-    label: string
-    cols: number
-    gap: number
-    children: DiagramNode[]
-    fill?: string
-    stroke?: string
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-}
-
-/**
- * A swimlane pool: a sparse grid of (lane, column) cells.
- *
- * `lanes` names the role bands. Each child declares which cell it occupies, and empty
- * cells stay empty — that is the whole point of a swimlane diagram, where a step belongs
- * to exactly one role and the columns show the order things happen in.
- *
- * `phases` is an optional band of milestone labels above the columns.
- */
-export interface PoolNode {
-    kind: "pool"
-    id: string
-    label: string
-    /** Role names, one per band. */
-    lanes: string[]
-    /** Milestone labels spanning the columns. Empty means no milestone band. */
-    phases: string[]
-    /** "horizontal": lanes stack downwards, flow left to right. "vertical": the mirror. */
-    orientation: "horizontal" | "vertical"
-    gap: number
-    children: DiagramNode[]
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-}
-
-/**
- * A sequence diagram: participants across the top, lifelines hanging below them.
- *
- * Children are the participant heads, in left-to-right order. The messages are ordinary
- * links whose `step` gives the vertical order — so the same `link` operation that draws
- * an arrow in a flowchart draws a message here.
- *
- * The engine emits the lifelines as separate cells; they are not nodes, because nothing
- * ever attaches to a lifeline directly.
- */
-export interface SequenceNode {
-    kind: "sequence"
-    id: string
-    label: string
-    /** Horizontal distance between participant centres. */
-    gap: number
-    /** Vertical distance between consecutive messages. */
-    step: number
-    children: DiagramNode[]
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-}
-
-/**
- * A mind map or org chart: a root with branches radiating from it.
- *
- * Children are a FLAT list of every node in the map. The hierarchy comes from the links —
- * an arrow from A to B means B is a branch of A — not from nesting.
- *
- * That is not a shortcut, it is the only thing that works: a branch of a mind map is a
- * labelled box that also has sub-branches, and a box cannot hold children. Reading the
- * hierarchy from the arrows also matches what the diagram means, since in a mind map or an
- * org chart the arrows ARE the structure.
- *
- * `spread: "radial"` fans branches out on both sides of the centre, which is what a mind
- * map wants. `spread: "down"` puts every branch below the centre, which is what an org
- * chart wants: a reporting line only reads correctly downwards.
- */
-export interface RadialNode {
-    kind: "radial"
-    id: string
-    label: string
-    spread: "radial" | "down"
-    /** Distance from a parent's edge to its children. */
-    gap: number
-    children: DiagramNode[]
-    style?: string
-    pinned?: boolean
-    rect?: Rect
-}
-
-export type ContainerNode =
-    | GroupNode
-    | GridNode
-    | PoolNode
-    | SequenceNode
-    | RadialNode
-export type LeafNode = IconNode | BoxNode | TitleNode
-export type DiagramNode = ContainerNode | LeafNode
-
-export interface Rect {
-    x: number
-    y: number
-    w: number
-    h: number
-}
-
-/** An arrow. Routing is the engine's business; the model only says what connects to what. */
-export interface LinkSpec {
-    /** Cell id, so an existing edge can be addressed by later operations. */
-    id?: string
-    source: string
-    target: string
-    label?: string
-    /** Dashed line — replication, sync, policy, lineage. */
-    dashed?: boolean
-    /**
-     * A bold arrow: the relationship IS the point — a transformation, the main flow.
-     * Thick and coloured, a visual element rather than a hairline connector.
-     */
-    bold?: boolean
-    /**
-     * Arrowhead at the target / at the source. draw.io endArrow/startArrow tokens:
-     * block, open, diamond, diamondThin, oval, cross, ERone, ERmany, ERoneToMany,
-     * ERzeroToMany, ERzeroToOne, none… Unset means the default (classic at the target,
-     * nothing at the source). `headFill`/`tailFill` distinguish UML composition
-     * (filled diamond) from aggregation (hollow) — conventions where fill IS meaning.
-     */
-    head?: string
-    tail?: string
-    headFill?: boolean
-    tailFill?: boolean
-    /** Step number, rendered as an "N. " prefix on the label. */
-    step?: number
-    /** Verbatim style, when recovered from XML. */
-    style?: string
-}
-
-/** A whole diagram page: the node forest plus its arrows. */
-export interface DiagramTree {
-    /** Top-level nodes, in layout order. */
-    roots: DiagramNode[]
-    links: LinkSpec[]
-    /** Page title, if the diagram has one. */
-    title?: string
-    /**
-     * Target width : height of the whole page. 1 is square, 1.6 landscape, 0.7 portrait.
-     *
-     * This is the one number that decides whether a diagram reads as a poster or as a
-     * letterbox, and it cannot be derived: the same content is a legitimate 1-column
-     * portrait or 3-column landscape. So the model declares it, the engine gives the top
-     * level a width to match, and every proportional rule below finally has a share of
-     * something real to divide up.
-     */
-    aspect?: number
-    /**
-     * Cells the parser could not fit into the tree — a user's own annotation boxes, a
-     * legend, shapes from an imported file. Kept verbatim and re-emitted untouched so
-     * a re-layout never destroys work the engine does not understand.
-     */
-    foreign: ForeignCell[]
-}
-
-/** A cell carried through the round-trip without interpretation. */
-export interface ForeignCell {
-    id: string
-    /** The cell's own serialised XML, verbatim. */
-    xml: string
-    /** Parent id at parse time, so it can be re-attached. */
-    parent: string
-}
-
-export function isContainer(n: DiagramNode): n is ContainerNode {
-    return (
-        n.kind === "group" ||
-        n.kind === "grid" ||
-        n.kind === "pool" ||
-        n.kind === "sequence" ||
-        n.kind === "radial"
-    )
-}
-
-export function isLeaf(n: DiagramNode): n is LeafNode {
-    return !isContainer(n)
-}
-
-/** Depth-first walk over a node and its descendants. */
-export function* walk(n: DiagramNode): Generator<DiagramNode> {
-    yield n
-    if (isContainer(n)) for (const c of n.children) yield* walk(c)
-}
-
-/** Every node in a tree, in document order. */
-export function* walkTree(t: DiagramTree): Generator<DiagramNode> {
-    for (const r of t.roots) yield* walk(r)
-}
-
-/** Find a node by id, or null. */
-export function findNode(t: DiagramTree, id: string): DiagramNode | null {
-    for (const n of walkTree(t)) if (n.id === id) return n
-    return null
-}
-
-/** The container holding `id`, or null when it is a root or absent. */
-export function findParent(t: DiagramTree, id: string): ContainerNode | null {
-    for (const n of walkTree(t)) {
-        if (!isContainer(n)) continue
-        if (n.children.some((c) => c.id === id)) return n
-    }
-    return null
-}

lib/diagram-engine/visgraph.ts

@@ -1,260 +0,0 @@
-/**
- * Obstacle-avoiding orthogonal routing: the orthogonal visibility graph, and A* over it.
- *
- * The router beside this file works by trying a list of candidate shapes — straight, an L, a
- * Z with its trunk in some lane — and keeping the first that is clear. That can only ever be
- * as good as the list, and a fixed list is not enough: measured over 250 generated
- * flowcharts, 347 arrows were drawn through a box that had nothing to do with them. Adding
- * shapes to the list moves the failures around rather than removing them.
- *
- * This is the complete alternative, from Wybrow, Marriott & Stuckey, "Orthogonal Connector
- * Routing" (Graph Drawing 2009) — the algorithm behind libavoid. Two ideas make it work:
- *
- *   1. THE GRID IS FINITE AND SUFFICIENT. Take the "interesting points": every obstacle
- *      corner and every connection point. Their x-coordinates and y-coordinates define a
- *      grid. The paper's observation, with proof: for any valid orthogonal route there is a
- *      route using only this grid that is no longer and has no more bends — shrink each
- *      segment onto the nearest grid line. So searching the grid loses nothing, and there is
- *      no resolution to tune. This is what a uniform pixel grid gets wrong in both
- *      directions at once: too coarse and it cannot fit through a narrow gap, too fine and
- *      the search explodes.
- *
- *   2. THE STATE INCLUDES THE DIRECTION OF ARRIVAL. Bends have to be paid for, and whether
- *      the next step is a bend depends on which way this one came in. So a search state is
- *      (point, incoming direction), not just (point). Without that the cost function cannot
- *      see bends at all.
- *
- * The heuristic is the one libavoid uses: Manhattan distance to the target plus the minimum
- * number of bends still needed, times the bend cost. It never overestimates — the remaining
- * path is at least the straight-line Manhattan distance, and it must contain at least that
- * many bends — so A* returns a cheapest route, not merely a route.
- *
- * Written from the paper's description rather than ported: the reference implementation is a
- * C++ library built for interactive re-routing, with incremental scanline updates and pin
- * management that a one-shot XML generator has no use for.
- */
-
-import type { Rect } from "./types"
-
-export interface Point {
-    x: number
-    y: number
-}
-
-/** Which way a path segment travels. Indices are used as array offsets. */
-const DIRS = [
-    { dx: 0, dy: -1 }, // 0 north
-    { dx: 1, dy: 0 }, // 1 east
-    { dx: 0, dy: 1 }, // 2 south
-    { dx: -1, dy: 0 }, // 3 west
-] as const
-
-/**
- * Cost of one bend, in pixels of path length.
- *
- * libavoid's default is 10. It has to be positive or the search has no reason to prefer a
- * straight line to a staircase of the same length, and the two look nothing alike.
- */
-const BEND_COST = 10
-
-/** Clearance kept around an obstacle, matching the router's own margin. */
-const MARGIN = 7
-
-/**
- * Does the segment from `p` to `q` pass through any obstacle?
- *
- * Obstacles are expanded by `MARGIN` first, so a route grazing a border counts as a hit —
- * an arrow drawn hard against a box reads as touching it.
- */
-function blocked(p: Point, q: Point, obstacles: Rect[]): boolean {
-    const lo = { x: Math.min(p.x, q.x), y: Math.min(p.y, q.y) }
-    const hi = { x: Math.max(p.x, q.x), y: Math.max(p.y, q.y) }
-    for (const r of obstacles) {
-        if (
-            lo.x < r.x + r.w + MARGIN &&
-            hi.x > r.x - MARGIN &&
-            lo.y < r.y + r.h + MARGIN &&
-            hi.y > r.y - MARGIN
-        )
-            return true
-    }
-    return false
-}
-
-/**
- * The minimum number of bends to get from `p`, travelling in direction `d`, to `t`.
- *
- * This is the table in the paper's Figure 2(a), as an arithmetic rule rather than sixteen
- * cases. Two independent questions: is the target ahead along the current axis, and is it
- * off to the side? Each answer costs bends, and they compose.
- */
-function bendsToTarget(p: Point, d: number, t: Point): number {
-    const { dx, dy } = DIRS[d]
-    // How far the target lies along the direction of travel, and across it.
-    const along = dx !== 0 ? (t.x - p.x) * dx : (t.y - p.y) * dy
-    const across = dx !== 0 ? t.y - p.y : t.x - p.x
-    if (across === 0) {
-        // Dead ahead: no bend. Directly behind: out and back, two bends.
-        return along >= 0 ? 0 : 2
-    }
-    // Off to the side: one bend if it is also ahead, two if it is behind.
-    return along > 0 ? 1 : 2
-}
-
-/**
- * A cheapest obstacle-free orthogonal path from `from` to `to`, or null if none exists.
- *
- * `startDir` and `endDir` are the directions the path must leave and arrive by — the side of
- * the shape each end attaches to. Constraining them is what stops an arrow leaving a box and
- * immediately turning back across it: a departure direction the search must honour on its
- * first step cannot double back.
- *
- * `extraLanes` lets the caller add grid lines the obstacles alone would not produce, which
- * matters when a port sits somewhere other than an obstacle corner.
- */
-export function routeOrthogonal(
-    from: Point,
-    to: Point,
-    startDir: number,
-    endDir: number,
-    obstacles: Rect[],
-    extraLanes: { xs: number[]; ys: number[] } = { xs: [], ys: [] },
-): Point[] | null {
-    // --- the interesting-points grid
-    const xs = new Set<number>([from.x, to.x, ...extraLanes.xs])
-    const ys = new Set<number>([from.y, to.y, ...extraLanes.ys])
-    for (const r of obstacles) {
-        // Just outside each edge, so a lane hugging an obstacle is still usable.
-        xs.add(r.x - MARGIN - 1)
-        xs.add(r.x + r.w + MARGIN + 1)
-        ys.add(r.y - MARGIN - 1)
-        ys.add(r.y + r.h + MARGIN + 1)
-    }
-    const X = [...xs].sort((a, b) => a - b)
-    const Y = [...ys].sort((a, b) => a - b)
-    const xi = new Map(X.map((v, i) => [v, i]))
-    const yi = new Map(Y.map((v, i) => [v, i]))
-
-    const sx = xi.get(from.x)
-    const sy = yi.get(from.y)
-    const tx = xi.get(to.x)
-    const ty = yi.get(to.y)
-    if (sx == null || sy == null || tx == null || ty == null) return null
-
-    // --- A* over (grid point, incoming direction)
-    const key = (ix: number, iy: number, d: number) =>
-        (iy * X.length + ix) * 4 + d
-    const best = new Map<number, number>()
-    const parent = new Map<number, number>()
-    // A binary heap would be tidier, but the frontier stays small on diagram-sized inputs and
-    // a sorted insert keeps this readable.
-    const open: { ix: number; iy: number; d: number; g: number; f: number }[] =
-        []
-    const push = (ix: number, iy: number, d: number, g: number, f: number) => {
-        let lo = 0
-        let hi = open.length
-        while (lo < hi) {
-            const mid = (lo + hi) >> 1
-            if (open[mid].f > f) lo = mid + 1
-            else hi = mid
-        }
-        open.splice(lo, 0, { ix, iy, d, g, f })
-    }
-
-    const h = (ix: number, iy: number, d: number) =>
-        Math.abs(X[ix] - to.x) +
-        Math.abs(Y[iy] - to.y) +
-        bendsToTarget({ x: X[ix], y: Y[iy] }, d, to) * BEND_COST
-
-    const startKey = key(sx, sy, startDir)
-    best.set(startKey, 0)
-    push(sx, sy, startDir, 0, h(sx, sy, startDir))
-
-    // The path must ARRIVE travelling in `endDir`, so that is the only accepting state.
-    const goalKey = key(tx, ty, endDir)
-    let found = false
-
-    while (open.length > 0) {
-        const cur = open.pop() as {
-            ix: number
-            iy: number
-            d: number
-            g: number
-            f: number
-        }
-        const ck = key(cur.ix, cur.iy, cur.d)
-        if (cur.g > (best.get(ck) ?? Number.POSITIVE_INFINITY)) continue
-        if (ck === goalKey) {
-            found = true
-            break
-        }
-
-        const here = { x: X[cur.ix], y: Y[cur.iy] }
-        for (let nd = 0; nd < 4; nd++) {
-            // No reversing: it can never help, and it lets a path retrace itself.
-            if (nd === (cur.d + 2) % 4) continue
-            const { dx, dy } = DIRS[nd]
-            // Step to the NEXT grid line in this direction — the grid's whole point is that
-            // intermediate positions cannot change whether a route is clear.
-            const nix = cur.ix + dx
-            const niy = cur.iy + dy
-            if (nix < 0 || nix >= X.length || niy < 0 || niy >= Y.length)
-                continue
-            const next = { x: X[nix], y: Y[niy] }
-            if (blocked(here, next, obstacles)) continue
-
-            const step = Math.abs(next.x - here.x) + Math.abs(next.y - here.y)
-            const g = cur.g + step + (nd === cur.d ? 0 : BEND_COST)
-            const nk = key(nix, niy, nd)
-            if (g >= (best.get(nk) ?? Number.POSITIVE_INFINITY)) continue
-            best.set(nk, g)
-            parent.set(nk, ck)
-            push(nix, niy, nd, g, g + h(nix, niy, nd))
-        }
-    }
-
-    if (!found) return null
-
-    // --- rebuild, then drop the points that are not bends
-    const path: Point[] = []
-    let node: number | undefined = goalKey
-    while (node !== undefined) {
-        const d = node % 4
-        const rest = (node - d) / 4
-        path.unshift({
-            x: X[rest % X.length],
-            y: Y[(rest - (rest % X.length)) / X.length],
-        })
-        node = parent.get(node)
-    }
-    return simplify(path)
-}
-
-/** Drop collinear and duplicate points: draw.io renders a redundant waypoint as a kink. */
-function simplify(pts: Point[]): Point[] {
-    const out: Point[] = []
-    for (const p of pts) {
-        const last = out[out.length - 1]
-        if (last && Math.abs(last.x - p.x) < 1 && Math.abs(last.y - p.y) < 1)
-            continue
-        out.push(p)
-    }
-    const kept: Point[] = []
-    for (let i = 0; i < out.length; i++) {
-        if (i === 0 || i === out.length - 1) {
-            kept.push(out[i])
-            continue
-        }
-        const prev = kept[kept.length - 1]
-        const next = out[i + 1]
-        const collinear =
-            (Math.abs(prev.x - out[i].x) < 1 &&
-                Math.abs(out[i].x - next.x) < 1) ||
-            (Math.abs(prev.y - out[i].y) < 1 && Math.abs(out[i].y - next.y) < 1)
-        if (!collinear) kept.push(out[i])
-    }
-    return kept
-}
-
-/** The direction leaving a given side of a shape: away from it. */
-export const SIDE_DIR = { T: 0, R: 1, B: 2, L: 3 } as const

lib/dynamo-quota-manager.ts

@@ -64,10 +64,13 @@ interface QuotaCheckResult {
  * Check all quotas and increment request count atomically.
  * Uses composite key (PK=user, SK=date) for per-day tracking.
  * Each day automatically gets a new item - no explicit reset needed.
+ * A request limit of 0 means none; increment 0 checks the limits without
+ * counting a request (the screenshot check).
  */
 export async function checkAndIncrementRequest(
     ip: string,
     limits: QuotaLimits,
+    increment = 1,
 ): Promise<QuotaCheckResult> {
     // Skip if quota tracking not enabled
     if (!client || !TABLE) {
@@ -99,7 +102,7 @@ export async function checkAndIncrementRequest(
                      attribute_not_exists(tpmCount) OR tpmCount < :tpmLimit)
                 `,
                 ExpressionAttributeValues: {
-                    ":one": { N: "1" },
+                    ":one": { N: String(increment) },
                     ":minute": { S: currentMinute },
                     ":reqLimit": { N: String(limits.requests || 999999) },
                     ":tokenLimit": { N: String(limits.tokens || 999999) },

lib/i18n/dictionaries/en.json

@@ -132,6 +132,8 @@
         "customSystemMessage": "Custom System Message",
         "customSystemMessageDescription": "Add custom instructions appended to the AI's system prompt.",
         "customSystemMessagePlaceholder": "e.g., Always use blue color scheme for diagrams...",
+        "maxOutputTokens": "Max Output Tokens",
+        "maxOutputTokensDescription": "Budget for one reply, shared by thinking and the diagram XML. Raise it if the AI keeps thinking and no diagram appears. Leave empty for the default.",
         "panelVisibility": "Lobby Panels",
         "panelVisibilityDescription": "Choose which panels to show on the chat lobby.",
         "showRecentChats": "Recent Chats",
@@ -176,9 +178,6 @@
         "networkError": "Network error. Please check your connection.",
         "retryLimit": "Auto-retry limit reached ({max}). Please try again manually.",
         "continuationRetryLimit": "Continuation retry limit reached ({max}). The diagram may be too complex.",
-        "validationFailed": "Diagram validation failed. Please try regenerating.",
-        "malformedXml": "AI generated invalid diagram XML. Please try regenerating.",
-        "failedToProcess": "Failed to process diagram. Please try regenerating.",
         "sessionCorrupted": "Session data was corrupted. Starting fresh.",
         "failedToSave": "Failed to save messages to localStorage",
         "failedToRestore": "Failed to restore from localStorage",
@@ -186,7 +185,26 @@
         "failedToExport": "Error fetching chart data",
         "failedToLoadExample": "Error loading example image",
         "failedToRecordFeedback": "Failed to record your feedback. Please try again.",
-        "storageUpdateFailed": "Chat cleared but browser storage could not be updated"
+        "storageUpdateFailed": "Chat cleared but browser storage could not be updated",
+        "sessionSaveFailed": "Could not save this chat. Browser storage may be full: delete old chats from history and try again.",
+        "sessionSaveFailedLeave": "Could not save this chat. Browser storage may be full. You can go on without saving it, then delete old chats from the list in the new chat.",
+        "continueWithoutSaving": "Continue without saving",
+        "llm": {
+            "invalid_api_key": "The provider rejected the API key. Check it in model settings.",
+            "forbidden": "The provider refused the request. The key may not have access to this model or region.",
+            "server_key_forbidden": "Today's free quota is used up. It resets tomorrow. You can also add your own API key in model settings to keep going.",
+            "model_not_found": "The provider does not know this model. Check the model ID in model settings.",
+            "insufficient_quota": "The provider account has no credit or quota left.",
+            "rate_limited": "The provider is limiting requests. Wait a moment and try again.",
+            "context_too_long": "The conversation is too long for this model. Start a new chat or pick a model with a larger context.",
+            "images_unsupported": "This model doesn't support image input.",
+            "tools_unsupported": "This model doesn't support tool calls, which drawing needs. Pick another model.",
+            "output_truncated": "The output was cut off before the diagram was complete. Try a simpler request or raise the output limit in settings.",
+            "provider_unavailable": "The provider is having problems. Try again later.",
+            "cannot_connect": "Could not reach the provider. Check the base URL and your network.",
+            "timeout": "The provider did not answer in time.",
+            "openModelSettings": "Open model settings"
+        }
     },
     "quota": {
         "dailyLimit": "Daily Quota Reached",
@@ -233,6 +251,7 @@
     "reasoning": {
         "thinking": "Thinking...",
         "thoughtFor": "Thought for {duration} seconds",
+        "thoughtForOne": "Thought for 1 second",
         "thoughtBrief": "Thought for a few seconds"
     },
     "dev": {
@@ -369,6 +388,11 @@
         "enterSecretKey": "Enter your secret access key",
         "baseUrl": "Base URL",
         "optional": "(optional)",
+        "getApiKey": "Get API key",
+        "fetchModels": "Fetch models from the provider",
+        "noTools": "no tool calls",
+        "mayNotDraw": "models.dev lists no tool call support for this model, so it may not be able to draw.",
+        "requestUrl": "Requests go to {url}",
         "baseUrlWithExample": "Base URL (optional, e.g. {example})",
         "customEndpoint": "Custom endpoint URL",
         "minimaxBaseUrlHint": "Use /anthropic for Anthropic-compatible API (recommended), or /v1 for OpenAI-compatible API",

lib/i18n/dictionaries/ja.json

@@ -132,6 +132,8 @@
         "customSystemMessage": "カスタムシステムメッセージ",
         "customSystemMessageDescription": "AIのシステムプロンプトに追加されるカスタム指示を入力します。",
         "customSystemMessagePlaceholder": "例：ダイアグラムには常に青色のカラースキームを使用...",
+        "maxOutputTokens": "最大出力トークン数",
+        "maxOutputTokensDescription": "1回の応答の予算で、思考過程とダイアグラムの XML が共有します。AI が考え続けてダイアグラムが生成されない場合は大きくしてください。空欄ならデフォルト値を使います。",
         "panelVisibility": "ロビーパネル",
         "panelVisibilityDescription": "チャットロビーに表示するパネルを選択します。",
         "showRecentChats": "最近のチャット",
@@ -176,9 +178,6 @@
         "networkError": "ネットワークエラー。接続を確認してください。",
         "retryLimit": "自動再試行制限に達しました（{max}）。手動で再試行してください。",
         "continuationRetryLimit": "継続再試行制限に達しました（{max}）。ダイアグラムが複雑すぎる可能性があります。",
-        "validationFailed": "ダイアグラムの検証に失敗しました。再生成してみてください。",
-        "malformedXml": "AI が無効なダイアグラム XML を生成しました。再生成してみてください。",
-        "failedToProcess": "ダイアグラムの処理に失敗しました。再生成してみてください。",
         "sessionCorrupted": "セッションデータが破損しました。最初からやり直します。",
         "failedToSave": "localStorage へのメッセージの保存に失敗しました",
         "failedToRestore": "localStorage からの復元に失敗しました",
@@ -186,7 +185,26 @@
         "failedToExport": "チャートデータの取得エラー",
         "failedToLoadExample": "例の画像の読み込みエラー",
         "failedToRecordFeedback": "フィードバックの記録に失敗しました。もう一度お試しください。",
-        "storageUpdateFailed": "チャットはクリアされましたが、ブラウザストレージを更新できませんでした"
+        "storageUpdateFailed": "チャットはクリアされましたが、ブラウザストレージを更新できませんでした",
+        "sessionSaveFailed": "このチャットを保存できませんでした。ブラウザのストレージがいっぱいの可能性があります。履歴から古いチャットを削除して、もう一度お試しください。",
+        "sessionSaveFailedLeave": "このチャットを保存できませんでした。ブラウザのストレージがいっぱいの可能性があります。保存せずに続けて、新しいチャットの一覧から古いチャットを削除できます。",
+        "continueWithoutSaving": "保存せずに続ける",
+        "llm": {
+            "invalid_api_key": "プロバイダーが API キーを拒否しました。モデル設定で確認してください。",
+            "forbidden": "プロバイダーがリクエストを拒否しました。このキーにはこのモデルまたはリージョンの利用権限がない可能性があります。",
+            "server_key_forbidden": "本日の無料枠を使い切りました。明日になると自動的に回復します。モデル設定でご自身の API キーを入力すると、引き続きご利用いただけます。",
+            "model_not_found": "プロバイダーがこのモデルを認識できません。モデル設定でモデル ID を確認してください。",
+            "insufficient_quota": "プロバイダーのアカウントの残高または利用枠がなくなりました。",
+            "rate_limited": "プロバイダーがリクエスト数を制限しています。少し待ってから再試行してください。",
+            "context_too_long": "会話がこのモデルで扱える長さを超えています。新しいチャットを始めるか、より長いコンテキストに対応したモデルを選んでください。",
+            "images_unsupported": "このモデルは画像入力に対応していません。",
+            "tools_unsupported": "このモデルはツール呼び出しに対応していません。作図にはツール呼び出しが必要です。別のモデルを選んでください。",
+            "output_truncated": "ダイアグラムが完成する前に出力が途中で切れました。リクエストを簡単にするか、設定で出力上限を上げてください。",
+            "provider_unavailable": "プロバイダーで問題が発生しています。しばらくしてから再試行してください。",
+            "cannot_connect": "プロバイダーに接続できません。Base URL とネットワークを確認してください。",
+            "timeout": "プロバイダーから時間内に応答がありませんでした。",
+            "openModelSettings": "モデル設定を開く"
+        }
     },
     "quota": {
         "dailyLimit": "1日の割当量に達しました",
@@ -233,6 +251,7 @@
     "reasoning": {
         "thinking": "考え中...",
         "thoughtFor": "{duration} 秒考えました",
+        "thoughtForOne": "1 秒考えました",
         "thoughtBrief": "数秒考えました"
     },
     "dev": {
@@ -323,6 +342,11 @@
         "enterSecretKey": "シークレットアクセスキーを入力",
         "baseUrl": "ベース URL",
         "optional": "（オプション）",
+        "getApiKey": "API キーを取得",
+        "fetchModels": "プロバイダーからモデル一覧を取得",
+        "noTools": "ツール呼び出し非対応",
+        "mayNotDraw": "models.dev によると、このモデルはツール呼び出しに対応していないため、作図できない可能性があります。",
+        "requestUrl": "リクエスト先: {url}",
         "baseUrlWithExample": "ベース URL（オプション、例: {example}）",
         "customEndpoint": "カスタムエンドポイント URL",
         "minimaxBaseUrlHint": "/anthropic で Anthropic 互換 API（推奨）、または /v1 で OpenAI 互換 API を使用",

lib/i18n/dictionaries/zh-Hant.json

@@ -132,6 +132,8 @@
         "customSystemMessage": "自訂系統訊息",
         "customSystemMessageDescription": "新增自訂指示，將附加到 AI 的系統提示末尾。",
         "customSystemMessagePlaceholder": "例如：圖表始終使用藍色配色方案...",
+        "maxOutputTokens": "最大輸出 token 數",
+        "maxOutputTokensDescription": "單次回覆的額度，思考過程與圖表 XML 共用。若 AI 一直在思考卻沒有產生圖表，請將它調大。留空則使用預設值。",
         "panelVisibility": "大廳面板",
         "panelVisibilityDescription": "選擇在聊天大廳顯示哪些面板。",
         "showRecentChats": "最近聊天",
@@ -176,9 +178,6 @@
         "networkError": "網路錯誤。請檢查您的連線。",
         "retryLimit": "已達自動重試限制（{max}）。請手動重試。",
         "continuationRetryLimit": "已達繼續重試限制（{max}）。圖表可能過於複雜。",
-        "validationFailed": "圖表驗證失敗。請嘗試重新產生。",
-        "malformedXml": "AI 產生的圖表 XML 無效。請嘗試重新產生。",
-        "failedToProcess": "無法處理圖表。請嘗試重新產生。",
         "sessionCorrupted": "工作階段資料已損壞。重新開始。",
         "failedToSave": "無法儲存訊息到 localStorage",
         "failedToRestore": "無法從 localStorage 還原",
@@ -186,7 +185,26 @@
         "failedToExport": "取得圖表資料時出錯",
         "failedToLoadExample": "載入範例圖片時出錯",
         "failedToRecordFeedback": "記錄您的回饋失敗。請重試。",
-        "storageUpdateFailed": "聊天已清除，但無法更新瀏覽器儲存空間"
+        "storageUpdateFailed": "聊天已清除，但無法更新瀏覽器儲存空間",
+        "sessionSaveFailed": "無法儲存這個對話。瀏覽器儲存空間可能已滿，請在歷史紀錄裡刪除舊對話後重試。",
+        "sessionSaveFailedLeave": "無法儲存這個對話，瀏覽器儲存空間可能已滿。可以不儲存它、直接繼續，再在新對話的列表裡刪除舊對話。",
+        "continueWithoutSaving": "不儲存，繼續",
+        "llm": {
+            "invalid_api_key": "服務商拒絕了這個 API Key，請在模型設定中檢查。",
+            "forbidden": "服務商拒絕了這次請求。這個 Key 可能沒有使用該模型或該地區的權限。",
+            "server_key_forbidden": "今天的免費額度已經用完，明天會自動恢復。您也可以在模型設定中填寫自己的 API Key 繼續使用。",
+            "model_not_found": "服務商找不到這個模型，請在模型設定中檢查模型 ID。",
+            "insufficient_quota": "服務商帳戶的餘額或額度已經用完。",
+            "rate_limited": "服務商正在限制請求頻率，請稍候再試。",
+            "context_too_long": "對話內容超過了這個模型能處理的長度。請開啟新的對話，或換一個上下文更長的模型。",
+            "images_unsupported": "這個模型不支援圖片輸入。",
+            "tools_unsupported": "這個模型不支援工具呼叫，而繪圖需要工具呼叫。請換一個模型。",
+            "output_truncated": "輸出在圖表完成之前就被截斷了。請簡化請求，或在設定中調高輸出上限。",
+            "provider_unavailable": "服務商發生問題，請稍後再試。",
+            "cannot_connect": "無法連線到服務商，請檢查 Base URL 和網路。",
+            "timeout": "服務商沒有及時回應。",
+            "openModelSettings": "開啟模型設定"
+        }
     },
     "quota": {
         "dailyLimit": "已達每日配額",
@@ -233,6 +251,7 @@
     "reasoning": {
         "thinking": "思考中...",
         "thoughtFor": "思考了 {duration} 秒",
+        "thoughtForOne": "思考了 1 秒",
         "thoughtBrief": "思考了幾秒鐘"
     },
     "dev": {
@@ -369,6 +388,11 @@
         "enterSecretKey": "輸入您的 Secret Key",
         "baseUrl": "基礎 URL",
         "optional": "（可選）",
+        "getApiKey": "取得 API Key",
+        "fetchModels": "從服務商取得模型清單",
+        "noTools": "不支援工具呼叫",
+        "mayNotDraw": "models.dev 顯示這個模型不支援工具呼叫，可能無法繪圖。",
+        "requestUrl": "請求將傳送至 {url}",
         "baseUrlWithExample": "基礎 URL（可選，例如 {example}）",
         "customEndpoint": "自訂端點 URL",
         "minimaxBaseUrlHint": "使用 /anthropic 端點為 Anthropic 相容 API（推薦），或使用 /v1 端點為 OpenAI 相容 API",

lib/i18n/dictionaries/zh.json

@@ -132,6 +132,8 @@
         "customSystemMessage": "自定义系统消息",
         "customSystemMessageDescription": "添加自定义指令，将附加到 AI 的系统提示末尾。",
         "customSystemMessagePlaceholder": "例如：图表始终使用蓝色配色方案...",
+        "maxOutputTokens": "最大输出 token 数",
+        "maxOutputTokensDescription": "单次回复的额度，思考过程和图表 XML 共用。如果 AI 一直在思考却没有生成图表，请把它调大。留空则使用默认值。",
         "panelVisibility": "大厅面板",
         "panelVisibilityDescription": "选择在聊天大厅显示哪些面板。",
         "showRecentChats": "最近聊天",
@@ -176,9 +178,6 @@
         "networkError": "网络错误。请检查您的连接。",
         "retryLimit": "已达到自动重试限制（{max}）。请手动重试。",
         "continuationRetryLimit": "已达到继续重试限制（{max}）。图表可能过于复杂。",
-        "validationFailed": "图表验证失败。请尝试重新生成。",
-        "malformedXml": "AI 生成的图表 XML 无效。请尝试重新生成。",
-        "failedToProcess": "无法处理图表。请尝试重新生成。",
         "sessionCorrupted": "会话数据已损坏。重新开始。",
         "failedToSave": "无法保存消息到 localStorage",
         "failedToRestore": "无法从 localStorage 恢复",
@@ -186,7 +185,26 @@
         "failedToExport": "获取图表数据时出错",
         "failedToLoadExample": "加载示例图片时出错",
         "failedToRecordFeedback": "记录您的反馈失败。请重试。",
-        "storageUpdateFailed": "聊天已清除，但无法更新浏览器存储"
+        "storageUpdateFailed": "聊天已清除，但无法更新浏览器存储",
+        "sessionSaveFailed": "无法保存这个对话。浏览器存储空间可能已满，请在历史记录里删除旧对话后重试。",
+        "sessionSaveFailedLeave": "无法保存这个对话，浏览器存储空间可能已满。可以不保存它、直接继续，再在新对话的列表里删除旧对话。",
+        "continueWithoutSaving": "不保存，继续",
+        "llm": {
+            "invalid_api_key": "服务商拒绝了这个 API Key，请在模型设置里检查。",
+            "forbidden": "服务商拒绝了这次请求。这个 Key 可能没有使用该模型或该地区的权限。",
+            "server_key_forbidden": "今天的免费额度已经用完，明天会自动恢复。您也可以在模型设置里填写自己的 API Key 继续使用。",
+            "model_not_found": "服务商找不到这个模型，请在模型设置里检查模型 ID。",
+            "insufficient_quota": "服务商账户的余额或额度已经用完。",
+            "rate_limited": "服务商正在限制请求频率，请稍等片刻再试。",
+            "context_too_long": "对话内容超过了这个模型能处理的长度。请新开一个对话，或换一个上下文更长的模型。",
+            "images_unsupported": "这个模型不支持图片输入。",
+            "tools_unsupported": "这个模型不支持工具调用，而画图需要工具调用。请换一个模型。",
+            "output_truncated": "输出在图画完之前就被截断了。请简化请求，或在设置里调高输出上限。",
+            "provider_unavailable": "服务商出了问题，请稍后再试。",
+            "cannot_connect": "连接不上服务商，请检查 Base URL 和网络。",
+            "timeout": "服务商没有及时响应。",
+            "openModelSettings": "打开模型设置"
+        }
     },
     "quota": {
         "dailyLimit": "已达每日配额",
@@ -233,6 +251,7 @@
     "reasoning": {
         "thinking": "思考中...",
         "thoughtFor": "思考了 {duration} 秒",
+        "thoughtForOne": "思考了 1 秒",
         "thoughtBrief": "思考了几秒钟"
     },
     "dev": {
@@ -369,6 +388,11 @@
         "enterSecretKey": "输入您的 Secret Key",
         "baseUrl": "基础 URL",
         "optional": "（可选）",
+        "getApiKey": "获取 API Key",
+        "fetchModels": "从服务商获取模型列表",
+        "noTools": "不支持工具调用",
+        "mayNotDraw": "models.dev 显示这个模型不支持工具调用，可能无法画图。",
+        "requestUrl": "请求将发往 {url}",
         "baseUrlWithExample": "基础 URL（可选，例如 {example}）",
         "customEndpoint": "自定义端点 URL",
         "minimaxBaseUrlHint": "使用 /anthropic 端点为 Anthropic 兼容 API（推荐），或使用 /v1 端点为 OpenAI 兼容 API",

lib/langfuse.ts

@@ -51,8 +51,15 @@ export function setTraceOutput(output: string) {
     if (!isLangfuseEnabled()) return
 
     updateActiveTrace({ output })
+    endTrace()
+}
+
+// End the observe() wrapper span (AI SDK creates its own child spans with usage).
+// It uses endOnExit: false, so every request path has to end it, or the trace
+// is never exported: stream finish, stream error/abort, and early returns.
+export function endTrace() {
+    if (!isLangfuseEnabled()) return
 
-    // End the observe() wrapper span (AI SDK creates its own child spans with usage)
     const activeSpan = api.trace.getActiveSpan()
     if (activeSpan) {
         activeSpan.end()

lib/llm-errors.ts

@@ -0,0 +1,195 @@
+import {
+    APICallError,
+    InvalidToolInputError,
+    LoadAPIKeyError,
+    NoSuchToolError,
+    RetryError,
+    ToolCallRepairError,
+} from "ai"
+
+/**
+ * What went wrong with a model call, for a hint the user can act on. The
+ * provider's own message always goes along, because a guess can be wrong.
+ */
+export type LLMErrorCode =
+    | "invalid_api_key"
+    | "forbidden"
+    | "model_not_found"
+    | "insufficient_quota"
+    | "rate_limited"
+    | "context_too_long"
+    | "images_unsupported"
+    | "tools_unsupported"
+    | "output_truncated"
+    | "provider_unavailable"
+    | "cannot_connect"
+    | "timeout"
+    // A 403 on the server's own key, e.g. a daily spend cap blocked it
+    | "server_key_forbidden"
+    | "unknown"
+
+export interface LLMError {
+    type: "provider"
+    code: LLMErrorCode
+    message: string
+}
+
+// Texts that name the cause more precisely than the status code: a quota
+// error can come as 403 or 429, a context or image error as a plain 400
+const SPECIFIC_TEXTS: Array<[RegExp, LLMErrorCode]> = [
+    [
+        // Not "too many tokens": that is Bedrock's throttling message
+        /context length|context window|maximum context|prompt is too long|input is too long|too many input tokens/i,
+        "context_too_long",
+    ],
+    [
+        /image content block|image_url|does not support image|image input is not supported/i,
+        "images_unsupported",
+    ],
+    [
+        /does not support tools|tool use is not supported|tools? (?:are|is) not supported|function calling is not supported/i,
+        "tools_unsupported",
+    ],
+    // Bedrock, when the output limit cut the tool call's JSON short
+    [/toolUse\.input is invalid/i, "output_truncated"],
+    // Bedrock, for a model id without the inference profile prefix
+    [/on-demand throughput isn.t supported/i, "model_not_found"],
+    [
+        /insufficient[_ ]quota|insufficient balance|exceeded your current quota|credit balance is too low|余额不足/i,
+        "insufficient_quota",
+    ],
+]
+
+const STATUS_CODES: Record<number, LLMErrorCode> = {
+    401: "invalid_api_key",
+    402: "insufficient_quota",
+    // Not "invalid key": a valid key can lack access to a model or region
+    403: "forbidden",
+    404: "model_not_found",
+    408: "timeout",
+    // A retired model
+    410: "model_not_found",
+    413: "context_too_long",
+    429: "rate_limited",
+}
+
+const GENERAL_TEXTS: Array<[RegExp, LLMErrorCode]> = [
+    [
+        /model[_ ]not[_ ]found|model .*does not exist|unknown model|no such model/i,
+        "model_not_found",
+    ],
+    [
+        /invalid[_ ]api[_ ]key|incorrect api key|unauthorized/i,
+        "invalid_api_key",
+    ],
+    // "too many tokens": Bedrock's throttling
+    [/rate limit|too many requests|too many tokens/i, "rate_limited"],
+    [
+        /Cannot connect to API|ECONNREFUSED|ENOTFOUND|ECONNRESET|ETIMEDOUT|fetch failed/i,
+        "cannot_connect",
+    ],
+]
+
+/** Secrets a provider may echo back: API keys, Bearer tokens, key=value */
+function redact(text: string): string {
+    return text
+        .replace(/\b(sk|pk|rk|ak)-[A-Za-z0-9_-]{8,}/g, "$1-[redacted]")
+        .replace(/\bBearer\s+[A-Za-z0-9._~+/-]+=*/gi, "Bearer [redacted]")
+        .replace(/\bAKIA[0-9A-Z]{16}\b/g, "[redacted]")
+        .replace(
+            /\b(api[_-]?key|access[_-]?key|secret|token|password|signature)(["']?\s*[:=]\s*["']?)[^\s"',&}]+/gi,
+            "$1$2[redacted]",
+        )
+}
+
+function problemDetail(body: string): string | undefined {
+    try {
+        const detail = JSON.parse(body)?.detail
+        return typeof detail === "string" ? detail : undefined
+    } catch {
+        return undefined
+    }
+}
+
+/**
+ * The error text for the chat stream: what went wrong with the provider as
+ * JSON for the hint, or the text the model must read to fix a tool call.
+ * On the server's keys the provider's own text stays in the server log:
+ * it can name the server's account, role or internal hosts.
+ */
+export function streamErrorText(error: unknown, hideDetails = false): string {
+    // The SDK passes an invalid tool call's error as a plain string. Other
+    // strings come from providers (DeepSeek's SDK sends stream errors so).
+    if (
+        typeof error === "string" &&
+        /^(Invalid input for tool|Model tried to call unavailable tool)/.test(
+            error,
+        )
+    ) {
+        return error
+    }
+    if (isToolCallError(error)) return (error as Error).message
+    const classified = classifyLLMError(error)
+    if (hideDetails) {
+        console.error("[chat] Provider error:", error)
+        if (classified.code === "forbidden") {
+            // The hint says all the user can do; there is no message to add
+            classified.code = "server_key_forbidden"
+            classified.message = ""
+        } else {
+            classified.message = "The provider returned an error."
+        }
+    }
+    return JSON.stringify(classified)
+}
+
+/**
+ * Model and tool errors the SDK sends back to the model as the tool result,
+ * so it can fix its call. Their text has to stay as it is.
+ */
+export function isToolCallError(error: unknown): boolean {
+    return (
+        InvalidToolInputError.isInstance(error) ||
+        NoSuchToolError.isInstance(error) ||
+        ToolCallRepairError.isInstance(error)
+    )
+}
+
+export function classifyLLMError(error: unknown): LLMError {
+    // After the SDK's retries, the last attempt says what happened
+    const e = RetryError.isInstance(error) ? error.lastError : error
+    // Errors sent inside the stream can be plain objects like OpenRouter's
+    // { code: 503, message }
+    const plain = e as {
+        message?: unknown
+        code?: unknown
+        statusCode?: number
+    }
+    const raw =
+        e instanceof Error
+            ? e.message
+            : typeof plain?.message === "string"
+              ? plain.message
+              : String(e)
+    const body = APICallError.isInstance(e) ? (e.responseBody ?? "") : ""
+    // A problem+json body names the reason the SDK left out (NVIDIA: "Gone")
+    const detail = problemDetail(body)
+    const message = redact(detail ? `${raw}: ${detail}` : raw).slice(0, 500)
+    const text = `${raw} ${body}`
+    const status = APICallError.isInstance(e)
+        ? e.statusCode
+        : (plain?.statusCode ??
+          (typeof plain?.code === "number" ? plain.code : undefined))
+
+    const find = (rules: Array<[RegExp, LLMErrorCode]>) =>
+        rules.find(([pattern]) => pattern.test(text))?.[1]
+    const code =
+        (e instanceof Error && e.name === "TimeoutError" && "timeout") ||
+        (LoadAPIKeyError.isInstance(e) && "invalid_api_key") ||
+        find(SPECIFIC_TEXTS) ||
+        (status && STATUS_CODES[status]) ||
+        (status && status >= 500 && "provider_unavailable") ||
+        find(GENERAL_TEXTS) ||
+        "unknown"
+    return { type: "provider", code, message }
+}

lib/model-catalog.ts

@@ -0,0 +1,45 @@
+import catalog from "@/lib/model-catalog.json"
+import type { ProviderName } from "@/lib/types/model-config"
+
+/**
+ * What models.dev knows about a model (scripts/update-model-catalog.mjs).
+ * Only used for hints: requests are sent the same way either way, since
+ * the data can be wrong or out of date.
+ */
+export interface ModelInfo {
+    tools: boolean
+    images: boolean
+    reasoning: boolean
+    context?: number
+    output?: number
+}
+
+const CATALOG = catalog as Record<string, Record<string, ModelInfo>>
+
+/**
+ * The entry for a model: an exact match ignoring case, else the longest id
+ * the model id starts with, followed by "-", ":" or ".". So
+ * claude-sonnet-4-5-20250929 finds claude-sonnet-4-5, but gpt-4 does not
+ * find gpt-4o.
+ */
+export function getModelInfo(
+    provider: ProviderName,
+    modelId: string,
+): ModelInfo | undefined {
+    const models = CATALOG[provider]
+    if (!models) return undefined
+    const wanted = modelId.trim().toLowerCase()
+    let best: string | undefined
+    for (const id of Object.keys(models)) {
+        const lower = id.toLowerCase()
+        if (lower === wanted) return models[id]
+        if (
+            wanted.startsWith(lower) &&
+            "-:.".includes(wanted[lower.length]) &&
+            lower.length > (best?.length ?? 0)
+        ) {
+            best = id
+        }
+    }
+    return best ? models[best] : undefined
+}

lib/output-token-limit.ts

@@ -0,0 +1,224 @@
+import { wrapLanguageModel } from "ai"
+
+type WrappedModel = ReturnType<typeof wrapLanguageModel>
+
+/**
+ * Default output budget for a chat turn.
+ *
+ * This has to cover thinking + prose + the tool call, because reasoning models
+ * spend it in that order. Measured on deepseek-v4-flash: refining an existing
+ * diagram burned 16000 tokens on thinking alone and the request ended with
+ * finishReason "length" before display_diagram was ever called (issue #924).
+ * 64000 leaves room for the plan and the XML in one turn.
+ */
+export const DEFAULT_MAX_OUTPUT_TOKENS = 64000
+
+/** Ceiling for the user-supplied override, to catch typos like an extra zero. */
+export const MAX_OUTPUT_TOKENS_LIMIT = 200000
+
+/**
+ * Below this a diagram cannot come out whole, so a retry would just produce
+ * truncated XML instead of the provider's error. Better to surface the error.
+ */
+const MIN_USABLE_OUTPUT_TOKENS = 1024
+
+/**
+ * Retry budget when a rejection names the budget parameter but no number we can
+ * read. It is the default from before 64000, which these providers ran with.
+ */
+const FALLBACK_OUTPUT_TOKENS = 16000
+
+/** Status codes that can carry a complaint about the requested budget. */
+const BUDGET_REJECTION_STATUSES = new Set([400, 422])
+
+function usableLimit(value: number): number | null {
+    return value >= MIN_USABLE_OUTPUT_TOKENS ? value : null
+}
+
+/** Message and body of an error that may be about the budget, or null. */
+export function rejectionText(error: unknown): string | null {
+    const err = error as {
+        message?: unknown
+        responseBody?: unknown
+        statusCode?: unknown
+    }
+
+    // An auth or rate-limit failure is not about the budget, so leave it alone.
+    if (
+        typeof err?.statusCode === "number" &&
+        !BUDGET_REJECTION_STATUSES.has(err.statusCode)
+    ) {
+        return null
+    }
+
+    const text = [
+        typeof err?.message === "string" ? err.message : "",
+        typeof err?.responseBody === "string" ? err.responseBody : "",
+    ].join(" ")
+
+    return text.trim() ? text : null
+}
+
+/**
+ * A budget this large exceeds what some models accept. Providers reject it with a
+ * 400 that names the real limit, so we parse the number out and retry once
+ * instead of failing the turn.
+ *
+ * Formats seen in the wild:
+ * - Bedrock: "The maximum tokens you requested exceeds the model limit of 4096."
+ * - OpenRouter: "This endpoint's maximum context length is 64000 tokens. However,
+ *   you requested about 64025 tokens (25 of text input, 64000 in the output)."
+ *   Note this one is an input+output ceiling, so the input has to be subtracted.
+ *   vLLM and SGLang send the same kind of ceiling, with the input written as
+ *   "6000 in the messages", "has 6000 input tokens" or "6000 tokens from the input".
+ * - Anthropic: "max_tokens: 200000 > 64000, which is the maximum allowed..."
+ * - OpenAI: "This model supports at most 16384 completion tokens"
+ * - Volcengine Ark: "The parameter `max_tokens` specified in the request are not
+ *   valid: integer above maximum value, expected a value <= 32768, but got 64000"
+ * - DashScope: "Range of max_tokens should be [1, 8192]"
+ *
+ * Every pattern names tokens explicitly. A generic one (an earlier draft matched
+ * "lower than N") would reinterpret unrelated failures, and retrying on a bogus
+ * number turns a readable error into an empty diagram.
+ */
+function readCeiling(text: string): number | null {
+    // Combined input+output ceiling: subtract the input the provider counted,
+    // plus a small margin because its estimate is approximate.
+    const context = text.match(/maximum context length (?:is|of) (\d+)/i)
+    if (context) {
+        const input =
+            text.match(/(\d+) of text input/i) ||
+            text.match(/(\d+) in the messages/i) ||
+            text.match(/(\d+) tokens from the input/i) ||
+            text.match(/(\d+) input tokens/i)
+        return Number(context[1]) - (input ? Number(input[1]) : 0) - 1024
+    }
+
+    const output =
+        text.match(/model limit of (\d+)/i) ||
+        text.match(/> (\d+), which is the maximum/i) ||
+        text.match(/at most (\d+) completion tokens/i) ||
+        text.match(/max_\w*tokens.*?expected a value (?:<=|\\u003c=) (\d+)/i) ||
+        text.match(/Range of max_tokens should be \[1,\s*(\d+)\]/i)
+
+    return output ? Number(output[1]) : null
+}
+
+/** The usable output ceiling named in a rejection, or null. */
+export function parseOutputTokenLimit(error: unknown): number | null {
+    const text = rejectionText(error)
+    const ceiling = text ? readCeiling(text) : null
+    return ceiling === null ? null : usableLimit(ceiling)
+}
+
+/**
+ * Thinking budget the provider adds on top of maxOutputTokens. Bedrock and
+ * Anthropic send maxOutputTokens + budgetTokens as max_tokens, so a ceiling in
+ * their rejection covers both.
+ */
+function thinkingBudget(providerOptions: unknown): number {
+    const options = providerOptions as
+        | {
+              bedrock?: {
+                  reasoningConfig?: { type?: string; budgetTokens?: unknown }
+              }
+              anthropic?: {
+                  thinking?: { type?: string; budgetTokens?: unknown }
+              }
+          }
+        | undefined
+    const config =
+        options?.bedrock?.reasoningConfig ?? options?.anthropic?.thinking
+    return config?.type === "enabled" && typeof config.budgetTokens === "number"
+        ? config.budgetTokens
+        : 0
+}
+
+/**
+ * The budget to retry with after a rejection, or null to surface the error.
+ */
+export function retryOutputTokens(
+    error: unknown,
+    params: { maxOutputTokens?: number; providerOptions?: unknown },
+): number | null {
+    const requested = params.maxOutputTokens
+    const text = rejectionText(error)
+    if (!requested || !text) return null
+
+    const ceiling = readCeiling(text)
+    if (ceiling !== null) {
+        // The ceiling applies to what was actually sent, thinking included,
+        // so the retry has to leave room for the thinking too.
+        const thinking = thinkingBudget(params.providerOptions)
+        if (ceiling >= requested + thinking) return null
+        return usableLimit(ceiling - thinking)
+    }
+
+    // Names the budget parameter, but in a format we cannot read a number from
+    if (/max_\w*tokens/i.test(text) && requested > FALLBACK_OUTPUT_TOKENS) {
+        return FALLBACK_OUTPUT_TOKENS
+    }
+    return null
+}
+
+/**
+ * Retry the stream once with a smaller budget when the provider rejects the
+ * requested one. Without this, raising the default breaks every model whose
+ * ceiling is below it (measured: bedrock claude-3-haiku 4096, nova-lite 10000,
+ * openrouter deepseek-r1 64000 shared with the input).
+ */
+export function withOutputTokenLimitFallback(
+    model: WrappedModel,
+): WrappedModel {
+    return wrapLanguageModel({
+        model,
+        middleware: {
+            specificationVersion: "v3",
+            async wrapStream({ doStream, params, model: inner }) {
+                try {
+                    return await doStream()
+                } catch (error) {
+                    const retry = retryOutputTokens(error, params)
+                    if (!retry) throw error
+
+                    console.warn(
+                        `[maxOutputTokens] ${params.maxOutputTokens} rejected, retrying with ${retry}`,
+                    )
+                    return await inner.doStream({
+                        ...params,
+                        maxOutputTokens: retry,
+                    })
+                }
+            },
+        },
+    })
+}
+
+function validBudget(value: string | null | undefined): number | null {
+    const parsed = Number(value)
+    return Number.isInteger(parsed) &&
+        parsed > 0 &&
+        parsed <= MAX_OUTPUT_TOKENS_LIMIT
+        ? parsed
+        : null
+}
+
+/**
+ * Resolve the output budget: user setting (sent as a header so it works in the
+ * desktop app too), then server env, then the default. Both sources go through
+ * the same validation, so a typo in either falls back instead of reaching the
+ * provider.
+ *
+ * On the server's credentials the user setting can only lower the server value,
+ * so MAX_OUTPUT_TOKENS keeps capping what the server pays for.
+ */
+export function resolveMaxOutputTokens(
+    headerValue: string | null,
+    usesServerCredentials: boolean,
+): number {
+    const header = validBudget(headerValue)
+    const server =
+        validBudget(process.env.MAX_OUTPUT_TOKENS) ?? DEFAULT_MAX_OUTPUT_TOKENS
+    if (header === null) return server
+    return usesServerCredentials ? Math.min(header, server) : header
+}

lib/pdf-utils.ts

@@ -1,4 +1,4 @@
-import { extractText, getDocumentProxy } from "unpdf"
+import { extractText } from "unpdf"
 
 // Maximum characters allowed for extracted text (configurable via env)
 const DEFAULT_MAX_EXTRACTED_CHARS = 150000 // 150k chars
@@ -14,6 +14,7 @@ const TEXT_EXTENSIONS = [
     ".json",
     ".csv",
     ".xml",
+    ".svg",
     ".html",
     ".css",
     ".js",
@@ -43,8 +44,10 @@ const TEXT_EXTENSIONS = [
  */
 export async function extractPdfText(file: File): Promise<string> {
     const buffer = await file.arrayBuffer()
-    const pdf = await getDocumentProxy(new Uint8Array(buffer))
-    const { text } = await extractText(pdf, { mergePages: true })
+    // Pass raw bytes so unpdf destroys the PDF document when it is done
+    const { text } = await extractText(new Uint8Array(buffer), {
+        mergePages: true,
+    })
     return text as string
 }
 

lib/provider-models.ts

@@ -0,0 +1,259 @@
+import { createGateway } from "ai"
+import { getModelInfo } from "@/lib/model-catalog"
+import { readLimitedBody } from "@/lib/read-limited-body"
+import {
+    normalizeBaseUrl,
+    ollamaApiUrl,
+    PROVIDER_INFO,
+    type ProviderName,
+} from "@/lib/types/model-config"
+
+/** A model a provider offers. tools is false when it cannot call tools. */
+export interface ListedModel {
+    id: string
+    tools?: boolean
+}
+
+export const AIHUBMIX_MODELS_ENDPOINT = "https://aihubmix.com/api/v1/models"
+
+export function canListModels(provider: ProviderName): boolean {
+    return (
+        Object.hasOwn(PROVIDER_INFO, provider) &&
+        !!PROVIDER_INFO[provider].modelList
+    )
+}
+
+// Models in OpenAI-style lists that are not for chat
+const NON_CHAT =
+    /(?:^|[-/_])(?:embed(?:ding)?s?|whisper|tts|transcribe|dall-e|moderation|rerank|realtime|sora)(?:$|[-/_])|gpt-image/i
+
+const NON_CHAT_AIHUBMIX_TYPES = new Set([
+    "embedding",
+    "image_generation",
+    "rerank",
+    "transcription",
+    "tts",
+    "video",
+])
+
+/** Chat model ids from AIHubMix's public model list */
+export function extractAihubmixModelIds(payload: unknown): string[] {
+    const data = (payload as { data?: unknown })?.data
+    if (!Array.isArray(data)) return []
+    const ids = new Set<string>()
+    for (const item of data) {
+        const record = item as { model_id?: unknown; types?: unknown }
+        if (typeof record?.model_id !== "string" || !record.model_id.trim()) {
+            continue
+        }
+        const types = new Set(
+            typeof record.types === "string"
+                ? record.types.split(",").map((t) => t.trim())
+                : [],
+        )
+        if (!types.has("llm")) continue
+        if ([...NON_CHAT_AIHUBMIX_TYPES].some((t) => types.has(t))) continue
+        ids.add(record.model_id.trim())
+    }
+    return [...ids]
+}
+
+/**
+ * An error this module wrote itself. Only these texts reach the caller:
+ * the base URL is the caller's and may be an internal address, so anything
+ * else (a parse error quoting the body, a network error naming a host)
+ * stays in the server log.
+ */
+export class ModelListError extends Error {
+    constructor(
+        message: string,
+        readonly statusCode?: number,
+    ) {
+        super(message)
+        this.name = "ModelListError"
+    }
+}
+
+const MAX_LIST_BYTES = 2 * 1024 * 1024
+
+/** A fetch that reads at most MAX_LIST_BYTES of each response */
+function sizeLimitedFetch(fetchFn: typeof fetch): typeof fetch {
+    return async (input, init) => {
+        // Ends a download that is too large (the Gateway SDK passes no
+        // signal of its own)
+        const download = new AbortController()
+        const signal = init?.signal
+            ? AbortSignal.any([init.signal, download.signal])
+            : download.signal
+        const response = await fetchFn(input, { ...init, signal })
+        const body = await readLimitedBody(response, MAX_LIST_BYTES)
+        if (body === null) {
+            download.abort()
+            throw new ModelListError("The model list is too large.")
+        }
+        // The body is already decoded and has its own length now
+        const headers = new Headers(response.headers)
+        headers.delete("content-encoding")
+        headers.delete("content-length")
+        // Some statuses must have no body at all
+        const noBody = [101, 204, 205, 304].includes(response.status)
+        return new Response(noBody ? null : body, {
+            status: response.status,
+            statusText: response.statusText,
+            headers,
+        })
+    }
+}
+
+/** GET a JSON list; a failed request carries its status for the error hint */
+async function getJson(
+    url: string,
+    headers: Record<string, string>,
+    fetchFn: typeof fetch,
+): Promise<any> {
+    const response = await fetchFn(url, {
+        headers,
+        signal: AbortSignal.timeout(15_000),
+    })
+    if (!response.ok) {
+        throw new ModelListError(
+            `The model list request failed (${response.status})`,
+            response.status,
+        )
+    }
+    const text = await response.text()
+    try {
+        return JSON.parse(text)
+    } catch {
+        throw new ModelListError("The model list was not valid JSON.")
+    }
+}
+
+/**
+ * Where to list from without the user's base URL: where chat goes then. For
+ * Ollama without a key that is the server's Ollama, else the SDK's local
+ * default; a local default in PROVIDER_INFO (SGLang's) only fills the
+ * settings form.
+ */
+function listFallbackUrl(provider: ProviderName, apiKey?: string): string {
+    if (provider === "ollama" && !apiKey) {
+        return process.env.OLLAMA_BASE_URL || "http://127.0.0.1:11434/api"
+    }
+    const url = PROVIDER_INFO[provider].defaultBaseUrl
+    return url?.startsWith("https://") ? url : ""
+}
+
+/**
+ * The provider's chat models, with tool support from the provider's own
+ * data or else models.dev. Only the client's key is used, so the server's
+ * keys never go to a URL the client chose.
+ */
+export async function listProviderModels(
+    provider: ProviderName,
+    { apiKey, baseUrl }: { apiKey?: string; baseUrl?: string },
+    unlimitedFetch: typeof fetch = fetch,
+): Promise<ListedModel[]> {
+    const fetchFn = sizeLimitedFetch(unlimitedFetch)
+    const base = normalizeBaseUrl(baseUrl || listFallbackUrl(provider, apiKey))
+    const bearer: Record<string, string> = apiKey
+        ? { Authorization: `Bearer ${apiKey}` }
+        : {}
+    let models: ListedModel[]
+
+    // AIHubMix has a public list, unless the user points to another
+    // endpoint, which is OpenAI-compatible
+    const style =
+        provider === "aihubmix" &&
+        baseUrl &&
+        !/^https:\/\/aihubmix\.com(\/v1)?$/.test(base)
+            ? "openai"
+            : PROVIDER_INFO[provider].modelList
+
+    switch (style) {
+        case "anthropic": {
+            const data = await getJson(
+                `${base}/models?limit=1000`,
+                {
+                    "x-api-key": apiKey ?? "",
+                    "anthropic-version": "2023-06-01",
+                },
+                fetchFn,
+            )
+            models = (data.data ?? []).map((m: { id: string }) => ({
+                id: m.id,
+            }))
+            break
+        }
+        case "google": {
+            // The key goes in a header: in the URL it would end up in logs
+            const data = await getJson(
+                `${base}/models?pageSize=1000`,
+                { "x-goog-api-key": apiKey ?? "" },
+                fetchFn,
+            )
+            models = (data.models ?? [])
+                .filter((m: { supportedGenerationMethods?: string[] }) =>
+                    m.supportedGenerationMethods?.includes("generateContent"),
+                )
+                .map((m: { name: string }) => ({
+                    id: m.name.replace(/^models\//, ""),
+                }))
+            break
+        }
+        case "ollama": {
+            const data = await getJson(
+                `${ollamaApiUrl(base)}/tags`,
+                bearer,
+                fetchFn,
+            )
+            models = (data.models ?? []).map((m: { name: string }) => ({
+                id: m.name,
+            }))
+            break
+        }
+        case "openrouter": {
+            const data = await getJson(`${base}/models`, bearer, fetchFn)
+            models = (data.data ?? []).map(
+                (m: { id: string; supported_parameters?: string[] }) => ({
+                    id: m.id,
+                    ...(m.supported_parameters && {
+                        tools: m.supported_parameters.includes("tools"),
+                    }),
+                }),
+            )
+            break
+        }
+        case "gateway": {
+            const { models: entries } = await createGateway({
+                ...(apiKey && { apiKey }),
+                ...(baseUrl && { baseURL: base }),
+                fetch: fetchFn,
+            }).getAvailableModels()
+            models = entries
+                .filter((m) => !m.modelType || m.modelType === "language")
+                .map((m) => ({ id: m.id }))
+            break
+        }
+        case "aihubmix": {
+            const data = await getJson(AIHUBMIX_MODELS_ENDPOINT, {}, fetchFn)
+            models = extractAihubmixModelIds(data).map((id) => ({ id }))
+            break
+        }
+        default: {
+            if (!base) {
+                throw new ModelListError(
+                    `${PROVIDER_INFO[provider].label} needs a base URL to list its models.`,
+                )
+            }
+            const data = await getJson(`${base}/models`, bearer, fetchFn)
+            models = (data.data ?? [])
+                .map((m: { id: string }) => ({ id: m.id }))
+                .filter((m: ListedModel) => !NON_CHAT.test(m.id))
+        }
+    }
+
+    return models.map((m) => ({
+        ...m,
+        tools: m.tools ?? getModelInfo(provider, m.id)?.tools,
+    }))
+}

lib/read-limited-body.ts

@@ -0,0 +1,32 @@
+/**
+ * Read a response body, giving up once it passes maxBytes, so a huge
+ * download from a URL the client chose can't exhaust server memory.
+ * Returns null when it is too large; the caller then aborts the request,
+ * which ends the download.
+ */
+export async function readLimitedBody(
+    response: Response,
+    maxBytes: number,
+): Promise<ArrayBuffer | null> {
+    if (Number(response.headers.get("content-length")) > maxBytes) {
+        return null
+    }
+    if (!response.body) return new ArrayBuffer(0)
+
+    const reader = response.body.getReader()
+    const chunks: Uint8Array[] = []
+    let total = 0
+    while (true) {
+        const { done, value } = await reader.read()
+        if (done) break
+        total += value.byteLength
+        if (total > maxBytes) {
+            // Not awaited: a copy of the body that Next.js keeps (its fetch
+            // dedupe) can hold the cancel back until it is read
+            reader.cancel().catch(() => {})
+            return null
+        }
+        chunks.push(value)
+    }
+    return new Blob(chunks as BlobPart[]).arrayBuffer()
+}