fix: older defects (batch C) and the second batch's review

Chats: - New Chat right after an answer saves that chat once. Saves run one at a time and read the chat on screen when their turn comes; a save scheduled for a chat that is no longer on screen is dropped. A chat whose id was still on its way to the URL no longer comes back after New Chat (the next answer went into it). - Crossing the 768 px breakpoint keeps the chat panel: a streaming answer, unsaved messages and attachments stay. The panel gets the sizes of each side, and a panel collapsed on desktop opens on mobile. - The chat's export waits for its own reply: an edit's history export still on its way no longer answers it with the older diagram, and two file saves at once no longer swap results. - A second edit in one answer is previewed on the first edit's result. - Stop also ends a running screenshot check; a chat that cannot be saved (storage full) can be left with "Continue without saving". - Small diagrams with shapes count as diagrams; the tool card no longer crashes on malformed operations. Quota and providers: - Requests that reach the server's own endpoints count toward the quota: EdgeOne (always its own endpoint now), a private base URL whatever key header is sent, keyless Ollama without a URL. With the quota on, a redirect is followed only to a public address. The output cap applies to these requests too. - Stop records the tokens of the steps that finished; the screenshot check counts its tokens without counting a request. - EdgeOne configured only by AI_PROVIDER works, also in the admin Test, which forwards the access code. Azure set up only in the admin panel works in chat. The Test sends a Bedrock session token. - The admin panel's Test of an entry without a URL uses the server's URL as the server does (no private address check for it); the admin panel no longer writes an Ollama URL. MCP server: - Write tools and start_session run one at a time, so two at once never drop each other's change; a cancelled call waiting its turn is skipped. get_diagram and export_diagram keep the session they started with. - Export to .drawio first gets the user's latest edits from the browser. - History thumbnails: one that arrives after the next AI write is dropped; a sync reply keeps the image; a version that changed only page settings is its own entry. - A diagram over the 10 MB limit is saved without its image, or the user is told to download it (the server now answers 413 instead of cutting the connection). - Labels holding text like id='1' or parent='1' are no longer read as attributes (a layer or a parent was deleted). A broken bare <mxGraphModel> file is refused. - After a sync reply the tab no longer sends its autosave copy again. Desktop and files: - A newer switch of the same preset is not rolled back by an older one that failed. .env values with escaped quotes are read whole. - MCP saved files: a file that could not be read stays protected while a folder without permission hides it, and is saved again once deleted. - The desktop app reports "no chats" only when the count was read and no model settings are stored.
2026-10-11 12:09:53 +08:00 · 2026-10-05 18:57:02 +09:00
parent c75f74a5a0
commit c0fa997186
67 changed files with 3152 additions and 531 deletions
@@ -257,12 +257,7 @@ export function deriveEnvUpdates(
            if (p.baseUrl) updates.GOOGLE_VERTEX_BASE_URL = p.baseUrl
        } else if (p.provider === "ollama") {
            if (p.apiKey) updates.OLLAMA_API_KEY = p.apiKey
-            // A key without a URL is an Ollama Cloud key, as its Test sends
-            // it; chat sends a server key to OLLAMA_BASE_URL or local Ollama
-            if (p.baseUrl || p.apiKey) {
-                updates.OLLAMA_BASE_URL =
-                    p.baseUrl || PROVIDER_INFO.ollama.defaultBaseUrl || null
-            }
+            if (p.baseUrl) updates.OLLAMA_BASE_URL = p.baseUrl
        } else {
            const env = credEnvNames(p.provider, index)
            if (env.key && p.apiKey) updates[env.key] = p.apiKey
@@ -21,6 +21,7 @@ import {
    adminProvidersToConfig,
    loadAdminProviders,
 } from "@/lib/admin/providers"
+import { getApiEndpoint } from "@/lib/base-path"
 import { redirectGuardedFetch } from "@/lib/ssrf-protection"
 import {
    normalizeBaseUrl,
@@ -100,6 +101,9 @@ export interface ClientOverrides {
    awsSessionToken?: string | null
    // Vertex AI config
    vertexApiKey?: string | null // Express Mode API key
+    // baseUrl is the server's own <P>_BASE_URL (the admin panel's Test),
+    // not one a user chose: no redirect guard
+    trustedBaseUrl?: boolean
    // Custom headers (e.g., for EdgeOne cookie auth)
    headers?: Record<string, string>
    // Custom env var name(s) for server models
@@ -569,6 +573,7 @@ function detectProvider(): ProviderName | null {
 function validateProviderCredentials(
    provider: ProviderName,
    customApiKeyEnv?: string | string[],
+    customBaseUrlEnv?: string,
 ): void {
    // Handle array of env var names - at least one must be set
    if (Array.isArray(customApiKeyEnv)) {
@@ -604,9 +609,12 @@ function validateProviderCredentials(
        }
    }

-    // Azure requires either AZURE_BASE_URL or AZURE_RESOURCE_NAME in addition to API key
+    // Azure requires either AZURE_BASE_URL or AZURE_RESOURCE_NAME in addition
+    // to API key, or a server model's own URL variable (an admin panel entry)
    if (provider === "azure") {
-        const hasBaseUrl = !!process.env.AZURE_BASE_URL
+        const hasBaseUrl =
+            !!process.env.AZURE_BASE_URL ||
+            !!(customBaseUrlEnv && process.env[customBaseUrlEnv])
        const hasResourceName = !!process.env.AZURE_RESOURCE_NAME
        if (!hasBaseUrl && !hasResourceName) {
            throw new Error(
@@ -879,13 +887,20 @@ export function getAIModel(clientOverrides?: ClientOverrides): ModelConfig {

    // Only validate server credentials if client isn't providing their own API key
    if (!isClientOverride) {
-        validateProviderCredentials(provider, overrides?.apiKeyEnv)
+        validateProviderCredentials(
+            provider,
+            overrides?.apiKeyEnv,
+            overrides?.baseUrlEnv,
+        )
    }

    console.log(`[AI Provider] Initializing ${provider} with model: ${modelId}`)

    // Requests to a base URL the client chose must not follow redirects
-    const guardedFetch = overrides?.baseUrl ? redirectGuardedFetch() : undefined
+    const guardedFetch =
+        overrides?.baseUrl && !overrides.trustedBaseUrl
+            ? redirectGuardedFetch()
+            : undefined
    // Build provider-specific options from environment variables
    let providerOptions = buildProviderOptions(provider, modelId)
    let model: LanguageModel
@@ -1091,20 +1106,30 @@ export function getAIModel(clientOverrides?: ClientOverrides): ModelConfig {
    return { model, providerOptions, modelId, provider }
 }

+/**
+ * The deployment's EdgeOne Pages function, as an absolute URL (the SDK
+ * needs one), under the deployment's base path
+ */
+export function edgeOneEndpoint(req: Request): string {
+    const origin = req.headers.get("origin") || new URL(req.url).origin
+    return `${origin}${getApiEndpoint("/api/edgeai")}`
+}
+
 /**
 * The server's <P>_BASE_URL for a provider, which getAIModel uses for a
 * server model without a URL variable of its own (an admin panel entry
- * without a URL). Bedrock, EdgeOne and Ollama (the panel writes
- * OLLAMA_BASE_URL itself) have none.
+ * without a URL). None for Bedrock and EdgeOne, and none for Ollama and
+ * Vertex AI, whose variables the panel writes itself (before a save they
+ * still hold the entry's previous URL).
 */
 export function globalBaseUrl(provider: ProviderName): string | undefined {
-    if (["bedrock", "edgeone", "ollama"].includes(provider)) return undefined
+    if (["bedrock", "edgeone", "ollama", "vertexai"].includes(provider)) {
+        return undefined
+    }
    const name =
-        provider === "vertexai"
-            ? "GOOGLE_VERTEX_BASE_URL"
-            : provider === "gateway"
-              ? "AI_GATEWAY_BASE_URL"
-              : `${provider.toUpperCase()}_BASE_URL`
+        provider === "gateway"
+            ? "AI_GATEWAY_BASE_URL"
+            : `${provider.toUpperCase()}_BASE_URL`
    return process.env[name] || undefined
 }

@@ -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) },
@@ -187,6 +187,8 @@
        "failedToRecordFeedback": "Failed to record your feedback. Please try again.",
        "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.",
@@ -187,6 +187,8 @@
        "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",
        "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。",
@@ -187,6 +187,8 @@
        "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",
        "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。",
@@ -187,6 +187,8 @@
        "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",
        "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。",
@@ -199,13 +199,18 @@ export async function deleteSession(id: string): Promise<void> {
 }

 export async function getSessionCount(): Promise<number> {
-    if (!isIndexedDBAvailable()) return 0
+    return (await readSessionCount()) ?? 0
+}
+
+/** The number of saved chats, or null when it could not be read */
+export async function readSessionCount(): Promise<number | null> {
+    if (!isIndexedDBAvailable()) return null
    try {
        const db = await getDB()
        return await db.count(STORE_NAME)
    } catch (error) {
        console.error("Failed to get session count:", error)
-        return 0
+        return null
    }
 }

@@ -118,24 +118,52 @@ export function allowPrivateUrls(): boolean {

 /** A redirect the guard below refused; its text is safe to show */
 export class RedirectRefusedError extends Error {
-    constructor() {
-        super("Redirects are not allowed for custom base URLs")
+    constructor(message = "Redirects are not allowed for custom base URLs") {
+        super(message)
        this.name = "RedirectRefusedError"
    }
 }

+const MAX_REDIRECTS = 5
+
 /**
 * A fetch for requests to a base URL the client chose. With private URLs
 * blocked, a public URL could still redirect the request to an internal
- * host, so redirects are refused. Undefined when private URLs are allowed.
+ * host, so redirects are refused. With private URLs allowed but the quota
+ * on (DYNAMODB_QUOTA_TABLE), a request to a private address counts as the
+ * server's: redirects are followed only to public addresses, or a public
+ * URL could reach the server's own network uncounted. Undefined otherwise.
 */
 export function redirectGuardedFetch(): typeof fetch | undefined {
-    if (allowPrivateUrls()) return undefined
+    const blockAll = !allowPrivateUrls()
+    if (!blockAll && !process.env.DYNAMODB_QUOTA_TABLE) return undefined
    return async (input, init) => {
-        const response = await fetch(input, { ...init, redirect: "manual" })
-        if (response.status >= 300 && response.status < 400) {
-            throw new RedirectRefusedError()
+        let url = input instanceof Request ? input.url : String(input)
+        let next = init
+        for (let hop = 0; hop <= MAX_REDIRECTS; hop++) {
+            const response = await fetch(url, { ...next, redirect: "manual" })
+            const location = response.headers.get("location")
+            if (response.status < 300 || response.status >= 400 || !location) {
+                return response
+            }
+            if (blockAll) throw new RedirectRefusedError()
+            url = new URL(location, url).toString()
+            if (await isPrivateUrl(url)) {
+                throw new RedirectRefusedError(
+                    "Redirects to private addresses are not allowed",
+                )
+            }
+            // As fetch itself does: 303, and 301 or 302 after a POST, go on
+            // as a GET without the body
+            const method = (next?.method ?? "GET").toUpperCase()
+            if (
+                response.status === 303 ||
+                ((response.status === 301 || response.status === 302) &&
+                    method === "POST")
+            ) {
+                next = { ...next, method: "GET", body: undefined }
+            }
        }
-        return response
+        throw new RedirectRefusedError("Too many redirects")
    }
 }
@@ -1,6 +1,7 @@
 import { type ClassValue, clsx } from "clsx"
 import * as pako from "pako"
 import { twMerge } from "tailwind-merge"
+import { hasCells } from "@/packages/mcp-server/src/pages.ts"

 export function cn(...inputs: ClassValue[]) {
    return twMerge(clsx(inputs))
@@ -17,12 +18,14 @@ export function cn(...inputs: ClassValue[]) {
 export const MIN_REAL_DIAGRAM_LENGTH = 300

 /**
- * Check if diagram XML represents a real diagram (not just empty template).
+ * Check if diagram XML represents a real diagram (not just empty template):
+ * it has a shape (however short), or is long enough to hold pages worth
+ * keeping.
 * @param xml - The diagram XML string to check
 * @returns true if the XML is a real diagram with content
 */
 export function isRealDiagram(xml: string | undefined | null): boolean {
-    return !!xml && xml.length > MIN_REAL_DIAGRAM_LENGTH
+    return !!xml && (hasCells(xml) || xml.length > MIN_REAL_DIAGRAM_LENGTH)
 }

 // ============================================================================