fix(ollama): keep DeepSeek cloud thinking enabled

steipete · steipete · commit a1b89317a8bb · 2026-05-10T01:49:43.000+01:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -12,6 +12,7 @@ Docs: https://docs.openclaw.ai
 
 ### Fixes
 
+- Ollama: keep DeepSeek V4 cloud models thinking-capable even when Ollama Cloud `/api/show` omits the `thinking` capability, so `/think high` no longer rejects `ollama/deepseek-v4-*:cloud`.
 - ACP sessions: map canonical runtime options to backend-advertised ACP config keys like Claude's `effort` while keeping persisted OpenClaw state canonical. (#79926) Thanks @InTheCloudDan.
 - Gateway/watch: rebuild or restage missing bundled-plugin dist and runtime-postbuild outputs before launching the Gateway from a source checkout, preventing incomplete watch-mode runtime trees. (#70805) Thanks @rubencu.
 - CLI/update: allow restart health probes from the previous gateway protocol during self-update, and make plugin dry-runs report exact npm target versions instead of `unknown` while preserving unchanged status.
diff --git a/extensions/ollama/index.test.ts b/extensions/ollama/index.test.ts
@@ -27,18 +27,22 @@ const ensureOllamaModelPulledMock = vi.hoisted(() => vi.fn(async () => {}));
 const buildOllamaProviderMock = vi.hoisted(() => vi.fn());
 const queryOllamaModelShowInfoMock = vi.hoisted(() => vi.fn());
 const buildOllamaModelDefinitionMock = vi.hoisted(() =>
-  vi.fn((modelId: string, contextWindow?: number, capabilities?: string[]) => ({
-    id: modelId,
-    name: modelId,
-    reasoning: capabilities?.includes("thinking") ?? false,
-    input: capabilities?.includes("vision") ? ["text", "image"] : ["text"],
-    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-    contextWindow: contextWindow ?? 8192,
-    maxTokens: 8192,
-    compat: capabilities
-      ? { supportsTools: capabilities.includes("tools"), supportsUsageInStreaming: true }
-      : { supportsUsageInStreaming: true },
-  })),
+  vi.fn((modelId: string, contextWindow?: number, capabilities?: string[]) => {
+    const normalized = modelId.trim().toLowerCase();
+    const isKnownCloudReasoningModel = /^deepseek-v4-(?:flash|pro):cloud$/.test(normalized);
+    return {
+      id: modelId,
+      name: modelId,
+      reasoning: isKnownCloudReasoningModel || (capabilities?.includes("thinking") ?? false),
+      input: capabilities?.includes("vision") ? ["text", "image"] : ["text"],
+      cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: contextWindow ?? 8192,
+      maxTokens: 8192,
+      compat: capabilities
+        ? { supportsTools: capabilities.includes("tools"), supportsUsageInStreaming: true }
+        : { supportsUsageInStreaming: true },
+    };
+  }),
 );
 const createConfiguredOllamaStreamFnMock = vi.hoisted(() =>
   vi.fn((_params: { model: unknown; providerBaseUrl?: string }) => ({}) as never),
@@ -468,7 +472,7 @@ describe("ollama plugin", () => {
     });
     queryOllamaModelShowInfoMock.mockResolvedValueOnce({
       contextWindow: 1048576,
-      capabilities: ["completion", "tools", "thinking"],
+      capabilities: ["completion", "tools"],
     });
 
     try {
diff --git a/extensions/ollama/src/provider-models.test.ts b/extensions/ollama/src/provider-models.test.ts
@@ -269,6 +269,19 @@ describe("ollama provider models", () => {
     expect(textModel.compat?.supportsTools).toBe(true);
     expect(textModel.compat?.supportsUsageInStreaming).toBe(true);
 
+    const deepseekCloudModel = buildOllamaModelDefinition("deepseek-v4-pro:cloud", 1048576, [
+      "completion",
+      "tools",
+    ]);
+    expect(deepseekCloudModel.reasoning).toBe(true);
+    expect(deepseekCloudModel.compat?.supportsTools).toBe(true);
+
+    const deepseekCloudModelWithoutCapabilities = buildOllamaModelDefinition(
+      "deepseek-v4-flash:cloud",
+      1048576,
+    );
+    expect(deepseekCloudModelWithoutCapabilities.reasoning).toBe(true);
+
     const noCapabilities = buildOllamaModelDefinition("unknown-model", 65536);
     expect(noCapabilities.input).toEqual(["text"]);
     expect(noCapabilities.compat?.supportsUsageInStreaming).toBe(true);
diff --git a/extensions/ollama/src/provider-models.ts b/extensions/ollama/src/provider-models.ts
@@ -236,6 +236,11 @@ export function isReasoningModelHeuristic(modelId: string): boolean {
   return /r1|reasoning|think|reason/i.test(modelId);
 }
 
+function isKnownOllamaCloudReasoningModel(modelId: string): boolean {
+  const normalized = modelId.trim().toLowerCase();
+  return /^deepseek-v4-(?:flash|pro):cloud$/.test(normalized);
+}
+
 export function buildOllamaModelDefinition(
   modelId: string,
   contextWindow?: number,
@@ -244,9 +249,10 @@ export function buildOllamaModelDefinition(
   const hasVision = capabilities?.includes("vision") ?? false;
   const input: ("text" | "image")[] = hasVision ? ["text", "image"] : ["text"];
   const reasoning =
-    capabilities === undefined
+    isKnownOllamaCloudReasoningModel(modelId) ||
+    (capabilities === undefined
       ? isReasoningModelHeuristic(modelId)
-      : capabilities.includes("thinking");
+      : capabilities.includes("thinking"));
   const compat =
     capabilities === undefined
       ? { supportsUsageInStreaming: true }
