feat: async task pipeline for media and llm workflows

server/_core/llm.test.ts

@@ -68,6 +68,29 @@ describe("invokeLLM", () => {
     expect(JSON.parse(request.body)).not.toHaveProperty("thinking");
   });
 
+  it("allows overriding the model per request", async () => {
+    process.env.LLM_API_URL = "https://one.hao.work/v1/chat/completions";
+    process.env.LLM_API_KEY = "test-key";
+    process.env.LLM_MODEL = "qwen3.5-plus";
+
+    const fetchMock = vi.fn().mockResolvedValue({
+      ok: true,
+      json: async () => mockSuccessResponse,
+    });
+    vi.stubGlobal("fetch", fetchMock);
+
+    const { invokeLLM } = await import("./llm");
+    await invokeLLM({
+      model: "qwen3-vl-235b-a22b",
+      messages: [{ role: "user", content: "describe image" }],
+    });
+
+    const [, request] = fetchMock.mock.calls[0] as [string, { body: string }];
+    expect(JSON.parse(request.body)).toMatchObject({
+      model: "qwen3-vl-235b-a22b",
+    });
+  });
+
   it("falls back to legacy forge variables when LLM_* values are absent", async () => {
     delete process.env.LLM_API_URL;
     delete process.env.LLM_API_KEY;