fix(ollama): honor model requests and pull completion contracts (#117171)

Co-authored-by: Peter Steinberger <steipete@macos.shared>

extensions/ollama/provider-discovery.test.ts

@@ -215,11 +215,11 @@ describe("Ollama provider", () => {
       if (url.endsWith("/api/show")) {
         const rawBody = init?.body;
         const bodyText = typeof rawBody === "string" ? rawBody : "{}";
-        const parsed = JSON.parse(bodyText) as { name?: string };
-        if (parsed.name === "qwen3:32b") {
+        const parsed = JSON.parse(bodyText) as { model?: string };
+        if (parsed.model === "qwen3:32b") {
           return jsonResponse({ model_info: { "qwen3.context_length": 131072 } });
         }
-        if (parsed.name === "llama3.3:70b") {
+        if (parsed.model === "llama3.3:70b") {
           return jsonResponse({ model_info: { "llama.context_length": 65536 } });
         }
       }

extensions/ollama/src/node-inference.paired-node.e2e.test.ts

@@ -607,13 +607,7 @@ async function handleFakeOllamaRequest(
   }
   if (requestPath === "/api/show") {
     const body = await readRequestJson(request);
-    // Ollama documents `model`; the current provider sends its supported `name` alias.
-    const modelName =
-      typeof body.model === "string"
-        ? body.model
-        : typeof body.name === "string"
-          ? body.name
-          : undefined;
+    const modelName = typeof body.model === "string" ? body.model : undefined;
     if (!modelName) {
       response.statusCode = 400;
       response.end(JSON.stringify({ error: "model is required" }));

extensions/ollama/src/node-inference.test.ts

@@ -79,16 +79,16 @@ async function withOllamaServer<T>(
       return;
     }
     if (request.url === "/api/show") {
-      const body = (await readBody(request)) as { name?: string };
-      if (body.name) {
-        showRequests.push(body.name);
+      const body = (await readBody(request)) as { model?: string };
+      if (body.model) {
+        showRequests.push(body.model);
       }
-      if (body.name === "unknown:latest") {
+      if (body.model === "unknown:latest") {
         response.statusCode = 500;
         response.end(JSON.stringify({ error: "show failed" }));
         return;
       }
-      const embedding = body.name?.startsWith("embedding") === true;
+      const embedding = body.model?.startsWith("embedding") === true;
       response.end(
         JSON.stringify({
           capabilities: embedding ? ["embedding"] : ["completion", "tools"],

extensions/ollama/src/provider-models.test.ts

@@ -50,6 +50,18 @@ describe("ollama provider models", () => {
     expect(resolveOllamaApiBase("http://127.0.0.1:11434///")).toBe("http://127.0.0.1:11434");
   });
 
+  it("inspects local models using Ollama's canonical model request field", async () => {
+    const fetchMock = vi.fn(async (_input: string | URL | Request, _init?: RequestInit) =>
+      jsonResponse({ model_info: {} }),
+    );
+    vi.stubGlobal("fetch", fetchMock);
+
+    await readOllamaModelShowInfo("http://127.0.0.1:11434", "gemma4:e2b");
+
+    const request = fetchMock.mock.calls[0]?.[1] as RequestInit | undefined;
+    expect(JSON.parse(requestBodyText(request?.body))).toEqual({ model: "gemma4:e2b" });
+  });
+
   it("caps local discovered runtime context while preserving native metadata", () => {
     const provider = capLocalOllamaProviderContext({
       api: "ollama",
@@ -93,8 +105,8 @@ describe("ollama provider models", () => {
       if (!url.endsWith("/api/show")) {
         throw new Error(`Unexpected fetch: ${url}`);
       }
-      const body = JSON.parse(requestBodyText(init?.body)) as { name?: string };
-      if (body.name === "llama3:8b") {
+      const body = JSON.parse(requestBodyText(init?.body)) as { model?: string };
+      if (body.model === "llama3:8b") {
         return jsonResponse({ model_info: { "llama.context_length": 65536 } });
       }
       return jsonResponse({});
@@ -161,8 +173,8 @@ describe("ollama provider models", () => {
         });
       }
       if (url.endsWith("/api/show")) {
-        const body = JSON.parse(requestBodyText(init?.body)) as { name?: string };
-        const completion = body.name === "qwen-chat:latest";
+        const body = JSON.parse(requestBodyText(init?.body)) as { model?: string };
+        const completion = body.model === "qwen-chat:latest";
         return jsonResponse({
           capabilities: completion ? ["completion", "tools"] : ["embedding"],
           model_info: completion ? { "qwen.context_length": 32_768 } : {},
@@ -275,14 +287,14 @@ describe("ollama provider models", () => {
       if (!url.endsWith("/api/show")) {
         throw new Error(`Unexpected fetch: ${url}`);
       }
-      const body = JSON.parse(requestBodyText(init?.body)) as { name?: string };
-      if (body.name === "kimi-k2.5:cloud") {
+      const body = JSON.parse(requestBodyText(init?.body)) as { model?: string };
+      if (body.model === "kimi-k2.5:cloud") {
         return jsonResponse({
           model_info: { "kimi-k2.context_length": 262144 },
           capabilities: ["vision", "thinking", "completion", "tools"],
         });
       }
-      if (body.name === "glm-5.1:cloud") {
+      if (body.model === "glm-5.1:cloud") {
         return jsonResponse({
           model_info: { "glm5.context_length": 202752 },
           capabilities: ["thinking", "completion", "tools"],
@@ -409,7 +421,7 @@ describe("ollama provider models", () => {
     const model: OllamaTagModel = { name: "qwen3:32b", digest: "sha256:normalized-base" };
     const fetchMock = vi.fn(async (input: string | URL | Request, init?: RequestInit) => {
       expect(requestUrl(input)).toBe("http://127.0.0.1:11434/api/show");
-      expect(JSON.parse(requestBodyText(init?.body))).toEqual({ name: "qwen3:32b" });
+      expect(JSON.parse(requestBodyText(init?.body))).toEqual({ model: "qwen3:32b" });
       return jsonResponse({
         model_info: { "qwen3.context_length": 131072 },
         capabilities: ["thinking", "tools"],

extensions/ollama/src/provider-models.ts

@@ -172,7 +172,7 @@ export async function readOllamaModelShowInfo(
     init: {
       method: "POST",
       headers,
-      body: JSON.stringify({ name: modelName }),
+      body: JSON.stringify({ model: modelName }),
     },
     // Guard-owned timeoutMs also bounds DNS/proxy preflight; init.signal does not.
     timeoutMs: Math.min(opts?.timeoutMs ?? OLLAMA_SHOW_TIMEOUT_MS, OLLAMA_SHOW_TIMEOUT_MS),

extensions/ollama/src/setup-pull.test.ts

@@ -1,6 +1,6 @@
 import type { WizardPrompter } from "openclaw/plugin-sdk/setup";
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { pullOllamaModel } from "./setup-pull.js";
+import { pullOllamaModel, pullOllamaModelNonInteractive } from "./setup-pull.js";
 
 const fetchWithSsrFGuardMock = vi.hoisted(() => vi.fn());
 
@@ -17,6 +17,81 @@ describe("Ollama onboarding model pulls", () => {
     fetchWithSsrFGuardMock.mockReset();
   });
 
+  it("uses the canonical Ollama model request field and requires its success terminal", async () => {
+    const release = vi.fn(async () => {});
+    fetchWithSsrFGuardMock.mockResolvedValue({
+      response: new Response('{"status":"pulling manifest"}\n{"status":"success"}\n'),
+      release,
+    });
+    const progress = { update: vi.fn(), stop: vi.fn() };
+    const prompter = { progress: vi.fn(() => progress) } as unknown as WizardPrompter;
+
+    await expect(pullOllamaModel("http://127.0.0.1:11434", "gemma4:e2b", prompter)).resolves.toBe(
+      true,
+    );
+
+    const request = fetchWithSsrFGuardMock.mock.calls[0]?.[0] as { init?: { body?: string } };
+    expect(JSON.parse(request.init?.body ?? "null")).toEqual({ model: "gemma4:e2b" });
+    expect(progress.stop).toHaveBeenCalledWith("Downloaded gemma4:e2b");
+    expect(release).toHaveBeenCalledOnce();
+  });
+
+  it.each([
+    { label: "empty response", body: "" },
+    { label: "interrupted manifest download", body: '{"status":"pulling manifest"}\n' },
+    {
+      label: "interrupted model layer",
+      body: '{"status":"pulling abcdef123456","total":100,"completed":40}\n',
+    },
+    { label: "malformed stream", body: "not valid json\n" },
+    { label: "incomplete trailing record", body: '{"status":"success"' },
+  ])("does not report a completed model pull for an $label", async ({ body }) => {
+    const release = vi.fn(async () => {});
+    fetchWithSsrFGuardMock.mockResolvedValue({ response: new Response(body), release });
+    const progress = { update: vi.fn(), stop: vi.fn() };
+    const prompter = { progress: vi.fn(() => progress) } as unknown as WizardPrompter;
+
+    await expect(pullOllamaModel("http://127.0.0.1:11434", "gemma4:e2b", prompter)).resolves.toBe(
+      false,
+    );
+
+    expect(progress.stop).toHaveBeenCalledWith(
+      "Failed to download gemma4:e2b: pull stream ended before success",
+    );
+    expect(release).toHaveBeenCalledOnce();
+  });
+
+  it("accepts a final success record without a trailing newline", async () => {
+    fetchWithSsrFGuardMock.mockResolvedValue({
+      response: new Response('{"status":"success"}'),
+      release: vi.fn(async () => {}),
+    });
+    const progress = { update: vi.fn(), stop: vi.fn() };
+    const prompter = { progress: vi.fn(() => progress) } as unknown as WizardPrompter;
+
+    await expect(pullOllamaModel("http://127.0.0.1:11434", "gemma4:e2b", prompter)).resolves.toBe(
+      true,
+    );
+    expect(progress.stop).toHaveBeenCalledWith("Downloaded gemma4:e2b");
+  });
+
+  it("reports interrupted pulls as failures during non-interactive setup", async () => {
+    fetchWithSsrFGuardMock.mockResolvedValue({
+      response: new Response('{"status":"pulling manifest"}\n'),
+      release: vi.fn(async () => {}),
+    });
+    const runtime = { log: vi.fn(), error: vi.fn() };
+
+    await expect(
+      pullOllamaModelNonInteractive("http://127.0.0.1:11434", "gemma4:e2b", runtime as never),
+    ).resolves.toBe(false);
+
+    expect(runtime.error).toHaveBeenCalledWith(
+      "Failed to download gemma4:e2b: pull stream ended before success",
+    );
+    expect(runtime.log).not.toHaveBeenCalledWith("Downloaded gemma4:e2b");
+  });
+
   it("coerces non-Error stream failures through the shared error contract", async () => {
     const release = vi.fn(async () => {});
     const response = new Response(

extensions/ollama/src/setup-pull.ts

@@ -68,7 +68,7 @@ async function pullOllamaModelCore(params: {
       init: {
         method: "POST",
         headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ name: modelName }),
+        body: JSON.stringify({ model: modelName }),
       },
       signal: params.signal
         ? AbortSignal.any([responseController.signal, params.signal])
@@ -92,28 +92,25 @@ async function pullOllamaModelCore(params: {
       let pendingRecordBytes = 0;
       const layers = new Map<string, { total: number; completed: number }>();
 
-      const parseLine = (line: string): OllamaPullResult => {
-        const trimmed = line.trim();
-        if (!trimmed) {
-          return { ok: true };
+      const parseLine = (line: string): OllamaPullResult | undefined => {
+        if (!line.trim()) {
+          return undefined;
         }
         try {
-          const chunk = JSON.parse(trimmed) as OllamaPullChunk;
+          const chunk = JSON.parse(line) as OllamaPullChunk;
           if (chunk.error) {
             return { ok: false, message: `Download failed: ${chunk.error}` };
           }
-          if (!chunk.status) {
-            return { ok: true };
+          if (!chunk.status || chunk.status === "success") {
+            return chunk.status ? { ok: true } : undefined;
           }
           if (chunk.total && chunk.completed !== undefined) {
             layers.set(chunk.status, { total: chunk.total, completed: chunk.completed });
-            const totals = [...layers.values()].reduce(
-              (sum, layer) => ({
-                total: sum.total + layer.total,
-                completed: sum.completed + layer.completed,
-              }),
-              { total: 0, completed: 0 },
-            );
+            const totals = { total: 0, completed: 0 };
+            for (const layer of layers.values()) {
+              totals.total += layer.total;
+              totals.completed += layer.completed;
+            }
             params.onStatus?.(
               chunk.status,
               totals.total > 0 ? Math.round((totals.completed / totals.total) * 100) : null,
@@ -124,14 +121,18 @@ async function pullOllamaModelCore(params: {
         } catch {
           // Ignore malformed streaming lines from Ollama.
         }
-        return { ok: true };
+        return undefined;
       };
 
       try {
         for (;;) {
           const { done, value } = await readOllamaPullChunkWithIdleTimeout(reader);
           if (done) {
-            return parseLine(buffer);
+            const terminal = parseLine(buffer);
+            if (terminal) {
+              return terminal;
+            }
+            throw new Error("pull stream ended before success");
           }
           pendingRecordBytes = checkNdjsonRecordCap(value, pendingRecordBytes);
           buffer += decoder.decode(value, { stream: true });
@@ -139,7 +140,7 @@ async function pullOllamaModelCore(params: {
           buffer = lines.pop() ?? "";
           for (const line of lines) {
             const parsed = parseLine(line);
-            if (!parsed.ok) {
+            if (parsed) {
               return parsed;
             }
           }
@@ -154,8 +155,7 @@ async function pullOllamaModelCore(params: {
       await release();
     }
   } catch (err) {
-    const reason = formatErrorMessage(err);
-    return { ok: false, message: `Failed to download ${modelName}: ${reason}` };
+    return { ok: false, message: `Failed to download ${modelName}: ${formatErrorMessage(err)}` };
   } finally {
     clearTimeout(responseTimeout);
   }

extensions/ollama/src/setup.non-interactive-auth.test.ts

@@ -44,9 +44,11 @@ function createOllamaFetchMock(params: {
       return jsonResponse({ models: params.tags.map((name) => ({ name })) });
     }
     if (url.endsWith("/api/show")) {
-      const body = JSON.parse(requestBodyText(init?.body)) as { name?: string };
-      const contextWindow = body.name ? params.show?.[body.name] : undefined;
-      const capabilities = body.name ? (params.capabilities?.[body.name] ?? ["tools"]) : ["tools"];
+      const body = JSON.parse(requestBodyText(init?.body)) as { model?: string };
+      const contextWindow = body.model ? params.show?.[body.model] : undefined;
+      const capabilities = body.model
+        ? (params.capabilities?.[body.model] ?? ["tools"])
+        : ["tools"];
       return jsonResponse({
         ...(contextWindow ? { model_info: { "llama.context_length": contextWindow } } : {}),
         capabilities,
@@ -73,10 +75,21 @@ describe("Ollama non-interactive onboarding", () => {
     upsertAuthProfileWithLock.mockClear();
   });
 
-  it("does not persist local auth when non-interactive setup cannot select a model", async () => {
+  it.each([
+    {
+      label: "Ollama reports a pull failure",
+      body: '{"error":"disk full"}\n',
+      error: "Download failed: disk full",
+    },
+    {
+      label: "the model pull ends before success",
+      body: '{"status":"pulling manifest"}\n',
+      error: "Failed to download missing-model: pull stream ended before success",
+    },
+  ])("does not persist unavailable local models when $label", async ({ body, error }) => {
     const fetchMock = createOllamaFetchMock({
       tags: [],
-      pullResponse: new Response('{"error":"disk full"}\n', { status: 200 }),
+      pullResponse: new Response(body, { status: 200 }),
     });
     vi.stubGlobal("fetch", fetchMock);
     const runtime = createRuntime();
@@ -91,7 +104,7 @@ describe("Ollama non-interactive onboarding", () => {
       runtime,
     });
 
-    expect(runtime.error).toHaveBeenCalledWith("Download failed: disk full");
+    expect(runtime.error).toHaveBeenCalledWith(error);
     expect(runtime.error).toHaveBeenCalledWith(
       [
         "No Ollama models are available at http://127.0.0.1:11434.",
@@ -186,7 +199,7 @@ describe("Ollama non-interactive onboarding", () => {
           return false;
         }
         const init = call[1] as RequestInit | undefined;
-        return JSON.parse(requestBodyText(init?.body)).name === modelId;
+        return JSON.parse(requestBodyText(init?.body)).model === modelId;
       }),
     ).toHaveLength(1);
   });

extensions/ollama/src/setup.test.ts

@@ -56,12 +56,12 @@ function createOllamaFetchMock(params: {
       return jsonResponse({ models: (params.tags ?? []).map((name) => ({ name })) });
     }
     if (url.endsWith("/api/show")) {
-      const body = JSON.parse(requestBodyText(init?.body)) as { name?: string };
-      const contextWindow = body.name ? params.show?.[body.name] : undefined;
-      const capabilities = body.name
+      const body = JSON.parse(requestBodyText(init?.body)) as { model?: string };
+      const contextWindow = body.model ? params.show?.[body.model] : undefined;
+      const capabilities = body.model
         ? params.capabilities === undefined
           ? ["tools"]
-          : params.capabilities[body.name]
+          : params.capabilities[body.model]
         : undefined;
       return jsonResponse({
         ...(contextWindow ? { model_info: { "llama.context_length": contextWindow } } : {}),
@@ -569,7 +569,7 @@ describe("ollama setup", () => {
     });
     const pullCall = fetchMock.mock.calls.find((call) => requestUrl(call[0]).endsWith("/api/pull"));
     expect(pullCall).toBeDefined();
-    expect(JSON.parse(requestBodyText(pullCall?.[1]?.body))).toEqual({ name: "gemma4:e4b" });
+    expect(JSON.parse(requestBodyText(pullCall?.[1]?.body))).toEqual({ model: "gemma4:e4b" });
     expect(progress.update).toHaveBeenCalledWith("Downloading gemma4:e4b - pulling part - 50%");
     expect(progress.stop).toHaveBeenCalledWith("Downloaded gemma4:e4b");
     expect(result.config.models?.providers?.ollama?.models?.map((model) => model.id)).toContain(
@@ -657,7 +657,7 @@ describe("ollama setup", () => {
     const fetchMock = vi.fn(async (input: string | URL | Request, init?: RequestInit) => {
       if (requestUrl(input).endsWith("/api/show")) {
         const body = typeof init?.body === "string" ? JSON.parse(init.body) : {};
-        if (body.name === "broken:20b") {
+        if (body.model === "broken:20b") {
           return new Response("boom", { status: 500 });
         }
       }
@@ -714,8 +714,8 @@ describe("ollama setup", () => {
       markScanStarted = resolve;
     });
     const fetchMock = vi.fn(async (input: string | URL | Request, init?: RequestInit) => {
-      const body = init?.body ? (JSON.parse(requestBodyText(init.body)) as { name?: string }) : {};
-      if (!requestUrl(input).endsWith("/api/show") || body.name !== "model-200") {
+      const body = init?.body ? (JSON.parse(requestBodyText(init.body)) as { model?: string }) : {};
+      if (!requestUrl(input).endsWith("/api/show") || body.model !== "model-200") {
         return await baseFetch(input, init);
       }
       markScanStarted();
@@ -999,7 +999,7 @@ describe("ollama setup", () => {
     });
 
     const pullRequest = mockCallArg(fetchMock, 1, 1) as RequestInit | undefined;
-    expect(JSON.parse(requestBodyText(pullRequest?.body))).toEqual({ name: "llama3.2:latest" });
+    expect(JSON.parse(requestBodyText(pullRequest?.body))).toEqual({ model: "llama3.2:latest" });
     expect(result.agents?.defaults?.model).toEqual({ primary: "ollama/llama3.2:latest" });
     expect(upsertAuthProfileWithLock).toHaveBeenCalledTimes(1);
   });