Support Ollama model selection

apps/web/package.json

@@ -132,7 +132,7 @@
     "next-themes": "0.4.6",
     "nodemailer": "7.0.9",
     "nuqs": "2.7.2",
-    "ollama-ai-provider": "1.2.0",
+    "ollama-ai-provider-v2": "^1.5.5",
     "openai": "6.6.0",
     "p-queue": "9.0.0",
     "p-retry": "7.1.0",

apps/web/utils/actions/settings.validation.ts

@@ -38,7 +38,10 @@ export const saveAiSettingsBody = z
     aiApiKey: z.string().optional(),
   })
   .superRefine((val, ctx) => {
-    if (!val.aiApiKey && val.aiProvider !== DEFAULT_PROVIDER) {
+    const requiresApiKey =
+      val.aiProvider !== DEFAULT_PROVIDER && val.aiProvider !== Provider.OLLAMA;
+
+    if (!val.aiApiKey && requiresApiKey) {
       ctx.addIssue({
         code: z.ZodIssueCode.custom,
         message: "You must provide an API key for this provider",

apps/web/utils/llms/model.test.ts

@@ -3,6 +3,7 @@ import { getModel } from "./model";
 import { Provider } from "./config";
 import { env } from "@/env";
 import type { UserAIFields } from "./types";
+import { createOllama } from "ollama-ai-provider-v2";
 
 // Mock AI provider imports
 vi.mock("@ai-sdk/openai", () => ({
@@ -31,7 +32,7 @@ vi.mock("@openrouter/ai-sdk-provider", () => ({
   })),
 }));
 
-vi.mock("ollama-ai-provider", () => ({
+vi.mock("ollama-ai-provider-v2", () => ({
   createOllama: vi.fn(() => (model: string) => ({ model })),
 }));
 
@@ -50,6 +51,7 @@ vi.mock("@/env", () => ({
     ANTHROPIC_API_KEY: "test-anthropic-key",
     GROQ_API_KEY: "test-groq-key",
     OPENROUTER_API_KEY: "test-openrouter-key",
+    OPENROUTER_BACKUP_MODEL: "google/gemini-2.5-flash",
     OLLAMA_BASE_URL: "http://localhost:11434",
     NEXT_PUBLIC_OLLAMA_MODEL: "llama3",
     BEDROCK_REGION: "us-west-2",
@@ -75,6 +77,8 @@ describe("Models", () => {
     vi.mocked(env).DEFAULT_LLM_MODEL = undefined;
     vi.mocked(env).BEDROCK_ACCESS_KEY = "";
     vi.mocked(env).BEDROCK_SECRET_KEY = "";
+    vi.mocked(env).NEXT_PUBLIC_OLLAMA_MODEL = "llama3";
+    vi.mocked(env).OLLAMA_BASE_URL = "http://localhost:11434";
   });
 
   describe("getModel", () => {
@@ -153,18 +157,48 @@ describe("Models", () => {
       expect(result.model).toBeDefined();
     });
 
-    // it("should configure Ollama model correctly", () => {
-    //   const userAi: UserAIFields = {
-    //     aiApiKey: "user-api-key",
-    //     aiProvider: Provider.OLLAMA!,
-    //     aiModel: "llama3",
-    //   };
+    it("should configure Ollama model correctly", () => {
+      const userAi: UserAIFields = {
+        aiApiKey: null,
+        aiProvider: Provider.OLLAMA!,
+        aiModel: "llama3",
+      };
 
-    //   const result = getModel(userAi);
-    //   expect(result.provider).toBe(Provider.OLLAMA);
-    //   expect(result.modelName).toBe("llama3");
-    //   expect(result.model).toBeDefined();
-    // });
+      const result = getModel(userAi);
+      expect(result.provider).toBe(Provider.OLLAMA);
+      expect(result.modelName).toBe("llama3");
+      expect(result.model).toBeDefined();
+      expect(createOllama).toHaveBeenCalledWith({
+        baseURL: "http://localhost:11434",
+      });
+      expect(result.backupModel).toBeNull();
+    });
+
+    it("should throw when Ollama model is missing", () => {
+      const userAi: UserAIFields = {
+        aiApiKey: null,
+        aiProvider: Provider.OLLAMA!,
+        aiModel: null,
+      };
+
+      expect(() => getModel(userAi)).toThrow("Ollama model must be specified");
+    });
+
+    it("should fall back to default Ollama base URL when env missing", () => {
+      vi.mocked(env).OLLAMA_BASE_URL = undefined as any;
+
+      const userAi: UserAIFields = {
+        aiApiKey: null,
+        aiProvider: Provider.OLLAMA!,
+        aiModel: "llama3",
+      };
+
+      getModel(userAi);
+
+      expect(createOllama).toHaveBeenCalledWith({
+        baseURL: "http://localhost:11434",
+      });
+    });
 
     it("should configure Anthropic model correctly without Bedrock credentials", () => {
       const userAi: UserAIFields = {

apps/web/utils/llms/model.ts

@@ -6,7 +6,7 @@ import { createGoogleGenerativeAI } from "@ai-sdk/google";
 import { createGroq } from "@ai-sdk/groq";
 import { createOpenRouter } from "@openrouter/ai-sdk-provider";
 import { createGateway } from "@ai-sdk/gateway";
-// import { createOllama } from "ollama-ai-provider";
+import { createOllama } from "ollama-ai-provider-v2";
 import { env } from "@/env";
 import { Provider } from "@/utils/llms/config";
 import type { UserAIFields } from "@/utils/llms/types";
@@ -128,16 +128,21 @@ function selectModel(
       };
     }
     case Provider.OLLAMA: {
-      throw new Error(
-        "Ollama is not supported. Revert to version v1.7.28 or older to use it.",
-      );
-      // const modelName = aiModel || env.NEXT_PUBLIC_OLLAMA_MODEL;
-      // if (!modelName) throw new Error("Ollama model is not set");
-      // return {
-      //   provider: Provider.OLLAMA!,
-      //   modelName,
-      //   model: createOllama({ baseURL: env.OLLAMA_BASE_URL })(model),
-      // };
+      const modelName = aiModel;
+
+      if (!modelName) {
+        throw new Error("Ollama model must be specified");
+      }
+
+      const baseURL = env.OLLAMA_BASE_URL || "http://localhost:11434";
+      const ollama = createOllama({ baseURL });
+
+      return {
+        provider: Provider.OLLAMA!,
+        modelName,
+        model: ollama(modelName),
+        backupModel: null,
+      };
     }
 
     case Provider.BEDROCK: {
@@ -208,7 +213,7 @@ function createOpenRouterProviderOptions(
 function selectEconomyModel(userAi: UserAIFields): SelectModel {
   if (env.ECONOMY_LLM_PROVIDER && env.ECONOMY_LLM_MODEL) {
     const apiKey = getProviderApiKey(env.ECONOMY_LLM_PROVIDER);
-    if (!apiKey) {
+    if (!apiKey && providerRequiresApiKey(env.ECONOMY_LLM_PROVIDER)) {
       logger.warn("Economy LLM provider configured but API key not found", {
         provider: env.ECONOMY_LLM_PROVIDER,
       });
@@ -245,7 +250,7 @@ function selectEconomyModel(userAi: UserAIFields): SelectModel {
 function selectChatModel(userAi: UserAIFields): SelectModel {
   if (env.CHAT_LLM_PROVIDER && env.CHAT_LLM_MODEL) {
     const apiKey = getProviderApiKey(env.CHAT_LLM_PROVIDER);
-    if (!apiKey) {
+    if (!apiKey && providerRequiresApiKey(env.CHAT_LLM_PROVIDER)) {
       logger.warn("Chat LLM provider configured but API key not found", {
         provider: env.CHAT_LLM_PROVIDER,
       });
@@ -285,7 +290,7 @@ function selectDefaultModel(userAi: UserAIFields): SelectModel {
 
   // If user has not api key set, then use default model
   // If they do they can use the model of their choice
-  if (aiApiKey) {
+  if (aiApiKey || userAi.aiProvider === Provider.OLLAMA) {
     aiProvider = userAi.aiProvider || env.DEFAULT_LLM_PROVIDER;
     aiModel = userAi.aiModel || null;
   } else {
@@ -337,6 +342,10 @@ function getProviderApiKey(provider: string) {
   return providerApiKeys[provider];
 }
 
+function providerRequiresApiKey(provider: string) {
+  return provider !== Provider.OLLAMA;
+}
+
 function getBackupModel(userApiKey: string | null): LanguageModelV2 | null {
   // disable backup model if user is using their own api key
   if (userApiKey) return null;

docs/hosting/self-hosting.md

@@ -44,6 +44,20 @@ If doing this manually edit then you'll need to configure:
 - **LLM Provider**: Uncomment one provider block and add your API key
 - **Optional**: Microsoft OAuth, external Redis, etc.
 
+#### Using Ollama (local LLM)
+
+To use a locally hosted Ollama model instead of a cloud LLM provider:
+
+1. Set `NEXT_PUBLIC_OLLAMA_MODEL` in `apps/web/.env` to the exact model name you have pulled in Ollama (e.g., `llama3` or `qwen2.5`).
+2. (Optional) Set `OLLAMA_BASE_URL` if your Ollama server is not on the default `http://localhost:11434`. When running the app in Docker but Ollama is on the host, use `http://host.docker.internal:11434`.
+3. Restart the stack so the updated environment variables are loaded:
+
+```bash
+NEXT_PUBLIC_BASE_URL=https://yourdomain.com docker compose --env-file apps/web/.env --profile all up -d
+```
+
+No API key is required for Ollama. The UI will only show Ollama as a selectable provider when `NEXT_PUBLIC_OLLAMA_MODEL` is set.
+
 For detailed configuration instructions, see the [Environment Variables Reference](./environment-variables.md).
 
 **Note**: If you only want to use Microsoft and not Google OAuth then add skipped for the the Google client id and secret.