diff --git a/src/llm/provider/openai/openai-provider.test.ts b/src/llm/provider/openai/openai-provider.test.ts
index b4b6afda..783f87ad 100644
--- a/src/llm/provider/openai/openai-provider.test.ts
+++ b/src/llm/provider/openai/openai-provider.test.ts
@@ -262,3 +262,54 @@ describe("OpenAiProvider strictTools wiring", () => {
     });
   });
 });
+
+describe("OpenAiProvider without defaultChatModel", () => {
+  it("throws on complete() when defaultChatModel is not configured", async () => {
+    const p = new OpenAiProvider({
+      id: "test",
+      baseUrl: "https://example.invalid",
+      apiKey: "test-key",
+      fetchImpl: fakeFetch({
+        role: "assistant",
+        content: "ok",
+      }) as unknown as typeof fetch,
+    });
+    await expect(p.complete({ prompt: "hi" })).rejects.toThrow(
+      /has no defaultChatModel configured/,
+    );
+  });
+
+  it("throws on completeStream() when defaultChatModel is not configured", async () => {
+    const p = new OpenAiProvider({
+      id: "test",
+      baseUrl: "https://example.invalid",
+      apiKey: "test-key",
+      fetchImpl: fakeStreamFetch("ok") as unknown as typeof fetch,
+    });
+    const stream = p.completeStream({ prompt: "hi" });
+    await expect(stream.next()).rejects.toThrow(
+      /has no defaultChatModel configured/,
+    );
+  });
+
+  it("throws on describeImage() when defaultChatModel is not configured", async () => {
+    const p = new OpenAiProvider({
+      id: "test",
+      baseUrl: "https://example.invalid",
+      apiKey: "test-key",
+      supportsVision: true,
+      fetchImpl: fakeFetch({
+        role: "assistant",
+        content: "ok",
+      }) as unknown as typeof fetch,
+    });
+    await expect(
+      p.describeImage({
+        prompt: "describe",
+        images: [
+          { id: 0, bytes: new Uint8Array(), mimeType: "image/png" },
+        ],
+      }),
+    ).rejects.toThrow(/has no defaultChatModel configured/);
+  });
+});
diff --git a/src/llm/provider/openai/openai-provider.ts b/src/llm/provider/openai/openai-provider.ts
index 5712b3e3..9f3234a8 100644
--- a/src/llm/provider/openai/openai-provider.ts
+++ b/src/llm/provider/openai/openai-provider.ts
@@ -48,7 +48,7 @@ export interface OpenAiProviderOptions {
   id: string;
   baseUrl: string;
   apiKey: string;
-  defaultChatModel: string;
+  defaultChatModel?: string;
   headers?: Record<string, string>;
   /**
    * Header that carries the API key when the service does not accept
@@ -104,7 +104,7 @@ export class OpenAiProvider implements LlmProvider {
   readonly capabilities: ProviderCapabilities;
 
   private readonly http: OpenAiHttpDeps;
-  private readonly defaultChatModel: string;
+  private readonly defaultChatModel: string | undefined;
   private readonly apiPathPrefix: string;
   private readonly taggedToolCompatibility: "qwen" | undefined;
   private readonly extraBody: Record<string, unknown> | undefined;
@@ -157,18 +157,24 @@ export class OpenAiProvider implements LlmProvider {
   }
 
   async complete(request: CompletionRequest): Promise<CompletionResult> {
+    const model = this.defaultChatModel;
+    if (!model) {
+      throw new Error(
+        `openai-compatible provider "${this.name}" has no defaultChatModel configured`,
+      );
+    }
     // Unary only: sub-calls carry `response_format`, streamed turns never do.
     const json = await sendWithStructuredOutputFallback(
       {
         providerId: this.id,
-        model: this.defaultChatModel,
+        model,
         logger: this.http.logger,
       },
       request,
       (req) =>
         buildOpenAiChatBody(
           req,
-          this.defaultChatModel,
+          model,
           false,
           this.extraBody,
           this.maxOutputTokens,
@@ -187,15 +193,21 @@ export class OpenAiProvider implements LlmProvider {
       this.taggedToolCompatibility === "qwen"
         ? adaptQwenTaggedToolResponse(json, request)
         : json;
-    return normaliseOpenAiChatResponse(adapted, this.defaultChatModel);
+    return normaliseOpenAiChatResponse(adapted, model);
   }
 
   async *completeStream(
     request: CompletionRequest,
   ): AsyncGenerator<StreamChunk, CompletionResult, void> {
+    const model = this.defaultChatModel;
+    if (!model) {
+      throw new Error(
+        `openai-compatible provider "${this.name}" has no defaultChatModel configured`,
+      );
+    }
     const body = buildOpenAiChatBody(
       request,
-      this.defaultChatModel,
+      model,
       true,
       this.extraBody,
       this.maxOutputTokens,
@@ -334,7 +346,7 @@ export class OpenAiProvider implements LlmProvider {
     }
     const final = completionFromStreamFinal(
       streamFinal,
-      this.defaultChatModel,
+      model,
       accumulated,
       accumulatedReasoning,
     );
@@ -396,9 +408,15 @@ export class OpenAiProvider implements LlmProvider {
     if (!this.capabilities.vision) {
       throw new VisionUnsupportedError(this.name);
     }
+    const model = this.defaultChatModel;
+    if (!model) {
+      throw new Error(
+        `openai-compatible provider "${this.name}" has no defaultChatModel configured`,
+      );
+    }
     return describeImageViaOpenAi(
       this.http,
-      this.defaultChatModel,
+      model,
       request,
       this.apiPathPrefix,
       this.providerPreferences,
@@ -421,7 +439,7 @@ function normalizeApiPathPrefix(prefix: string): string {
 
 function completionFromStreamFinal(
   streamFinal: StreamFinalResult | void,
-  defaultChatModel: string,
+  defaultChatModel: string | undefined,
   accumulated: string,
   accumulatedReasoning: string,
 ): CompletionResult {
@@ -444,7 +462,7 @@ function completionFromStreamFinal(
     },
     cacheHitTokens: 0,
     slotId: -1,
-    modelId: streamFinal?.modelId ?? defaultChatModel,
+    modelId: streamFinal?.modelId ?? defaultChatModel ?? null,
     usage,
     toolCalls: streamFinal?.toolCalls,
     finishReason,
diff --git a/src/llm/provider/registry/provider-registry.test.ts b/src/llm/provider/registry/provider-registry.test.ts
index e7a67281..2e309421 100644
--- a/src/llm/provider/registry/provider-registry.test.ts
+++ b/src/llm/provider/registry/provider-registry.test.ts
@@ -87,4 +87,78 @@ describe("ProviderRegistry", () => {
       }),
     ).rejects.toThrow(/unknown llm provider kind/);
   });
+
+  it("allows embedding-only openai-compatible provider without defaultChatModel", async () => {
+    registerBuiltInProviderKinds();
+    const fakeConfig = {
+      ...getConfig(),
+      llm: {
+        activeTextProvider: "chat-provider",
+        activeEmbeddingProvider: "embed-only",
+        toolTransport: "auto" as const,
+        providers: [
+          {
+            id: "chat-provider",
+            kind: "openai-compatible",
+            baseUrl: "https://example.invalid",
+            defaultChatModel: "gpt-4",
+          },
+          {
+            id: "embed-only",
+            kind: "openai-compatible",
+            baseUrl: "https://example.invalid",
+            defaultEmbeddingModel: "nomic-embed-text",
+            userModels: [
+              {
+                id: "nomic-embed-text",
+                kind: "embedding" as const,
+                dim: 768,
+              },
+            ],
+          },
+        ],
+      },
+    } as AtomicAgentConfig;
+    const registry = await ProviderRegistry.fromConfig(fakeConfig, {
+      config: fakeConfig,
+      llamaClient: {} as never,
+      getProfile: () => ({}) as never,
+      logger: { debug: () => {}, info: () => {}, warn: () => {}, error: () => {} },
+    });
+    expect(registry.listIds()).toContain("embed-only");
+  });
+
+  it("rejects non-embedding openai-compatible provider without defaultChatModel", async () => {
+    registerBuiltInProviderKinds();
+    const fakeConfig = {
+      ...getConfig(),
+      llm: {
+        activeTextProvider: "chat-provider",
+        activeEmbeddingProvider: "embed-only",
+        toolTransport: "auto" as const,
+        providers: [
+          {
+            id: "chat-provider",
+            kind: "openai-compatible",
+            baseUrl: "https://example.invalid",
+            defaultChatModel: "gpt-4",
+          },
+          {
+            id: "bad-provider",
+            kind: "openai-compatible",
+            baseUrl: "https://example.invalid",
+            userModels: [{ id: "gpt-4", kind: "chat" as const }],
+          },
+        ],
+      },
+    } as AtomicAgentConfig;
+    await expect(
+      ProviderRegistry.fromConfig(fakeConfig, {
+        config: fakeConfig,
+        llamaClient: {} as never,
+        getProfile: () => ({}) as never,
+        logger: { debug: () => {}, info: () => {}, warn: () => {}, error: () => {} },
+      }),
+    ).rejects.toThrow(/requires defaultChatModel/);
+  });
 });
diff --git a/src/llm/provider/registry/register-built-in-providers.ts b/src/llm/provider/registry/register-built-in-providers.ts
index 1ef3d125..23d8654e 100644
--- a/src/llm/provider/registry/register-built-in-providers.ts
+++ b/src/llm/provider/registry/register-built-in-providers.ts
@@ -49,9 +49,18 @@ export function registerBuiltInProviderKinds(): void {
 
   registerProviderKind("openai-compatible", (ctx) => {
     const entry = ctx.entry;
-    if (!entry.baseUrl || !entry.defaultChatModel) {
+    const isEmbeddingOnly =
+      entry.userModels !== undefined &&
+      entry.userModels.length > 0 &&
+      entry.userModels.every((m) => m.kind === "embedding");
+    if (!entry.baseUrl) {
       throw new Error(
-        `openai-compatible provider "${entry.id}" requires baseUrl and defaultChatModel`,
+        `openai-compatible provider "${entry.id}" requires baseUrl`,
+      );
+    }
+    if (!isEmbeddingOnly && !entry.defaultChatModel) {
+      throw new Error(
+        `openai-compatible provider "${entry.id}" requires defaultChatModel`,
       );
     }
     return new OpenAiProvider({
@@ -73,9 +82,18 @@ export function registerBuiltInProviderKinds(): void {
 
   registerProviderKind("qwen-openai-compatible", (ctx) => {
     const entry = ctx.entry;
-    if (!entry.baseUrl || !entry.defaultChatModel) {
+    const isEmbeddingOnly =
+      entry.userModels !== undefined &&
+      entry.userModels.length > 0 &&
+      entry.userModels.every((m) => m.kind === "embedding");
+    if (!entry.baseUrl) {
+      throw new Error(
+        `qwen-openai-compatible provider "${entry.id}" requires baseUrl`,
+      );
+    }
+    if (!isEmbeddingOnly && !entry.defaultChatModel) {
       throw new Error(
-        `qwen-openai-compatible provider "${entry.id}" requires baseUrl and defaultChatModel`,
+        `qwen-openai-compatible provider "${entry.id}" requires defaultChatModel`,
       );
     }
     return new OpenAiProvider({