fix(ollama): send chat requests to the /api path

The Ollama SDK appends /chat to the base URL, and Ollama serves chat at
/api/chat. The model list already added the missing /api, so with
"http://localhost:11434" (the address our docs showed) models were listed
but every chat request went to /chat and got a 404. An OpenAI-style
".../v1" address went to /v1/chat.

ollamaApiUrl() turns the server address, ".../v1" and ".../api" into
".../api". Chat, the model list and the settings dialog's request URL
hint all use it. The docs now show http://localhost:11434/api, which also
works on released versions.

components/chat-panel.tsx

@@ -427,16 +427,13 @@ export default function ChatPanel({
             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
+                text = hints[data.code]
+                    ? `${hints[data.code]}\n\n${data.message}`
+                    : 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

docs/cn/ai-providers.md

@@ -192,7 +192,7 @@ 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

@@ -194,7 +194,7 @@ AI_MODEL=llama3.2
 Optional custom URL:
 
 ```bash
-OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_BASE_URL=http://localhost:11434/api
 ```
 
 ### ModelScope

docs/ja/ai-providers.md

@@ -179,7 +179,7 @@ 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

env.example

@@ -70,7 +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: Ollama Cloud; defaults to local Ollama (http://127.0.0.1:11434)
+# 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)
 

lib/ai-providers.ts

@@ -25,6 +25,7 @@ import { getEnvFallback } from "@/lib/admin/settings"
 import { isPrivateUrl, redirectGuardedFetch } from "@/lib/ssrf-protection"
 import {
     normalizeBaseUrl,
+    ollamaApiUrl,
     PROVIDER_INFO,
     type ProviderName,
 } from "@/lib/types/model-config"
@@ -1025,7 +1026,7 @@ export function getAIModel(clientOverrides?: ClientOverrides): ModelConfig {
                     ? PROVIDER_INFO.ollama.defaultBaseUrl
                     : resolveBaseUrlEnv(overrides, "OLLAMA_BASE_URL"))
             model = createOllama({
-                ...(baseURL && { baseURL }),
+                ...(baseURL && { baseURL: ollamaApiUrl(baseURL) }),
                 ...(apiKey && {
                     headers: { Authorization: `Bearer ${apiKey}` },
                 }),

lib/i18n/dictionaries/en.json

@@ -192,7 +192,6 @@
         "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.",

lib/i18n/dictionaries/ja.json

@@ -192,7 +192,6 @@
         "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。",

lib/i18n/dictionaries/zh-Hant.json

@@ -192,7 +192,6 @@
         "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。",

lib/i18n/dictionaries/zh.json

@@ -192,7 +192,6 @@
         "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。",

lib/llm-errors.ts

@@ -24,8 +24,6 @@ export type LLMErrorCode =
     | "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 {
@@ -132,13 +130,7 @@ export function streamErrorText(error: unknown, hideDetails = false): string {
     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."
-        }
+        classified.message = "The provider returned an error."
     }
     return JSON.stringify(classified)
 }

lib/provider-models.ts

@@ -3,6 +3,7 @@ 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"
@@ -200,8 +201,11 @@ export async function listProviderModels(
             break
         }
         case "ollama": {
-            const api = base.endsWith("/api") ? base : `${base}/api`
-            const data = await getJson(`${api}/tags`, bearer, fetchFn)
+            const data = await getJson(
+                `${ollamaApiUrl(base)}/tags`,
+                bearer,
+                fetchFn,
+            )
             models = (data.models ?? []).map((m: { name: string }) => ({
                 id: m.name,
             }))

lib/types/model-config.ts

@@ -622,6 +622,15 @@ export function normalizeBaseUrl(url: string): string {
         .replace(/\/(?:chat\/completions|completions|messages|responses)$/, "")
 }
 
+/**
+ * Ollama's native API root, which the SDK appends /chat to and the model
+ * list /tags. Users often enter the server address ("http://localhost:11434")
+ * or its OpenAI-compatible one (".../v1"); both get /api.
+ */
+export function ollamaApiUrl(baseUrl: string): string {
+    return `${normalizeBaseUrl(baseUrl).replace(/\/(?:api|v1)$/, "")}/api`
+}
+
 /** Where a chat request goes for a base URL, or null when the SDK decides */
 export function chatRequestUrl(
     provider: ProviderName,
@@ -630,6 +639,7 @@ export function chatRequestUrl(
     const url = normalizeBaseUrl(baseUrl)
     if (!url) return null
     if (provider === "anthropic") return `${url}/messages`
+    if (provider === "ollama") return `${ollamaApiUrl(url)}/chat`
     // These SDKs build their own paths (or, for MiniMax, pick the protocol
     // from the URL)
     const ownPaths: ProviderName[] = [
@@ -637,7 +647,6 @@ export function chatRequestUrl(
         "vertexai",
         "azure",
         "bedrock",
-        "ollama",
         "gateway",
         "minimax",
         "edgeone",

tests/e2e/chat-errors.spec.ts

@@ -75,30 +75,3 @@ test("a provider rate limit is not shown as this site's quota", async ({
     // The site's own tokens-per-minute toast
     await expect(page.getByText("Rate limit reached")).toHaveCount(0)
 })
-
-test("a refused server key shows only the quota hint and a settings button", async ({
-    page,
-}) => {
-    // What the chat route streams when the server's key gets a 403, e.g.
-    // after a daily spend cap blocked it
-    const errorText = JSON.stringify({
-        type: "provider",
-        code: "server_key_forbidden",
-        message: "",
-    })
-    await chatWith(page, {
-        status: 200,
-        contentType: "text/event-stream",
-        body: `data: {"type":"start"}\n\ndata: ${JSON.stringify({ type: "error", errorText })}\n\ndata: [DONE]\n\n`,
-    })
-    await expect(
-        page.getByText("Today's free quota is used up", { exact: false }),
-    ).toBeVisible({ timeout: 15000 })
-    await expect(page.getByText("The provider returned an error")).toHaveCount(
-        0,
-    )
-    await page.getByRole("button", { name: "Open model settings" }).click()
-    await expect(
-        page.getByRole("dialog", { name: "AI Model Configuration" }),
-    ).toBeVisible()
-})

tests/unit/ai-providers.test.ts

@@ -434,7 +434,7 @@ describe("Ollama API key security", () => {
 
         expect(createOllamaMock).toHaveBeenCalledWith(
             expect.objectContaining({
-                baseURL: "https://my-ollama.com",
+                baseURL: "https://my-ollama.com/api",
                 headers: { Authorization: "Bearer client-key" },
             }),
         )
@@ -448,7 +448,7 @@ describe("Ollama API key security", () => {
 
         expect(createOllamaMock).toHaveBeenCalledWith(
             expect.objectContaining({
-                baseURL: "https://cloud.ollama.com",
+                baseURL: "https://cloud.ollama.com/api",
                 headers: { Authorization: "Bearer server-key" },
             }),
         )
@@ -483,7 +483,24 @@ describe("Ollama API key security", () => {
 
         expect(createOllamaMock).toHaveBeenCalledTimes(1)
         const callArgs = createOllamaMock.mock.calls[0][0]
-        expect(callArgs.baseURL).toBe("https://my-ollama.com")
+        expect(callArgs.baseURL).toBe("https://my-ollama.com/api")
         expect(callArgs).not.toHaveProperty("headers")
     })
+
+    it("sends chat to Ollama's /api for a server address or a /v1 URL", () => {
+        delete process.env.OLLAMA_API_KEY
+
+        for (const baseUrl of [
+            "http://localhost:11434",
+            "http://localhost:11434/",
+            "http://localhost:11434/v1",
+            "http://localhost:11434/api",
+        ]) {
+            createOllamaMock.mockClear()
+            getAIModel({ provider: "ollama", baseUrl, modelId: "llama3.2" })
+            expect(createOllamaMock.mock.calls[0][0].baseURL).toBe(
+                "http://localhost:11434/api",
+            )
+        }
+    })
 })

tests/unit/base-url.test.ts

@@ -1,5 +1,9 @@
 import { describe, expect, it } from "vitest"
-import { chatRequestUrl, normalizeBaseUrl } from "@/lib/types/model-config"
+import {
+    chatRequestUrl,
+    normalizeBaseUrl,
+    ollamaApiUrl,
+} from "@/lib/types/model-config"
 
 describe("normalizeBaseUrl", () => {
     it("drops spaces, trailing slashes and a pasted endpoint path", () => {
@@ -32,6 +36,9 @@ describe("chatRequestUrl", () => {
         expect(
             chatRequestUrl("anthropic", "https://proxy.example.com/v1"),
         ).toBe("https://proxy.example.com/v1/messages")
+        expect(chatRequestUrl("ollama", "http://localhost:11434")).toBe(
+            "http://localhost:11434/api/chat",
+        )
     })
 
     it("stays out of the way for SDKs that build their own paths", () => {
@@ -40,3 +47,21 @@ describe("chatRequestUrl", () => {
         expect(chatRequestUrl("glm", "  ")).toBeNull()
     })
 })
+
+describe("ollamaApiUrl", () => {
+    it("points at Ollama's /api whatever form the address takes", () => {
+        for (const url of [
+            "http://localhost:11434",
+            "http://localhost:11434/",
+            "http://localhost:11434/api",
+            "http://localhost:11434/api/",
+            "http://localhost:11434/v1",
+            "http://localhost:11434/v1/chat/completions",
+        ]) {
+            expect(ollamaApiUrl(url)).toBe("http://localhost:11434/api")
+        }
+        expect(ollamaApiUrl("https://ollama.com/api")).toBe(
+            "https://ollama.com/api",
+        )
+    })
+})

tests/unit/chat-route-errors.test.ts

@@ -139,8 +139,7 @@ describe("provider error texts in the stream", () => {
         )
         const message = await streamedError({})
         expect(message).not.toMatch(/org-operator/)
-        // A 403 on the server's key gets its own hint and no message
-        expect(message).toBe("")
+        expect(message).toBe("The provider returned an error.")
     })
 })
 

tests/unit/llm-errors.test.ts

@@ -215,29 +215,11 @@ describe("streamErrorText", () => {
             "User: arn:aws:sts::123456789012:assumed-role/app/s is not authorized to perform: bedrock:InvokeModel",
         )
         const hidden = JSON.parse(streamErrorText(error, true))
+        expect(hidden.code).toBe("forbidden")
         expect(hidden.message).not.toMatch(/arn:aws|123456789012/)
         expect(JSON.parse(streamErrorText(error)).message).toMatch(
             /not authorized/,
         )
-        const throttled = JSON.parse(
-            streamErrorText(apiError(429, "Too many tokens"), true),
-        )
-        expect(throttled).toEqual({
-            type: "provider",
-            code: "rate_limited",
-            message: "The provider returned an error.",
-        })
-    })
-
-    it("names a 403 on the server's keys, e.g. a spend cap blocked them", () => {
-        const error = apiError(403, "explicit deny in an identity-based policy")
-        expect(JSON.parse(streamErrorText(error, true))).toEqual({
-            type: "provider",
-            code: "server_key_forbidden",
-            message: "",
-        })
-        // On the user's own key it stays a plain refusal
-        expect(JSON.parse(streamErrorText(error)).code).toBe("forbidden")
     })
 
     it("classifies a provider error", () => {