fix(inference): set compat.supportsUsageInStreaming for ollama-local (#2747)

src/lib/inference/config.test.ts

@@ -328,6 +328,43 @@ describe("getSandboxInferenceConfig", () => {
     });
   });
 
+  // #2747 followup to #3678: Ollama's OpenAI-compat stream omits the usage
+  // chunk unless the request sets `stream_options.include_usage: true`.
+  // OpenClaw gates that on `model.compat.supportsUsageInStreaming`. The
+  // ollama-local path runs through `inference.local`, which OpenClaw's
+  // own Ollama detector doesn't recognise — so the compat flag has to
+  // travel into openclaw.json via `inferenceCompat`. The previous Python-
+  // side conditional in `generate-openclaw-config.py` could never fire
+  // because `providerKey` resolves to `MANAGED_PROVIDER_ID` ("inference")
+  // for ollama-local, not "ollama" / "ollama-local". Wiring it here is
+  // the right layer.
+  it("forces supportsUsageInStreaming for ollama-local (#2747)", () => {
+    expect(getSandboxInferenceConfig("qwen2.5:7b", "ollama-local")).toEqual({
+      providerKey: MANAGED_PROVIDER_ID,
+      primaryModelRef: `${MANAGED_PROVIDER_ID}/qwen2.5:7b`,
+      inferenceBaseUrl: INFERENCE_ROUTE_URL,
+      inferenceApi: "openai-completions",
+      inferenceCompat: {
+        supportsUsageInStreaming: true,
+      },
+    });
+  });
+
+  it("does not force supportsUsageInStreaming for non-ollama providers", () => {
+    for (const provider of [
+      "nvidia-prod",
+      "nvidia-nim",
+      "nvidia-router",
+      "openai-api",
+      "compatible-endpoint",
+      "anthropic-prod",
+      "gemini-api",
+    ]) {
+      const compat = getSandboxInferenceConfig("model", provider).inferenceCompat;
+      expect(compat?.supportsUsageInStreaming).toBeUndefined();
+    }
+  });
+
   it("uses a probed Responses API override when one is available", () => {
     expect(getSandboxInferenceConfig("gpt-5.4", "openai-api", "openai-responses")).toEqual({
       providerKey: "openai",

src/lib/inference/config.ts

@@ -240,6 +240,23 @@
       providerKey = MANAGED_PROVIDER_ID;
       primaryModelRef = `${MANAGED_PROVIDER_ID}/${model}`;
       break;
+    case "ollama-local":
+      // Ollama's OpenAI-compatible /v1/chat/completions stream omits the
+      // `usage` chunk unless the request sets `stream_options.include_usage:
+      // true`. OpenClaw gates that on `model.compat.supportsUsageInStreaming`
+      // (src/agents/openai-transport-stream.ts upstream); its own Ollama
+      // extension auto-detects the flag only when the endpoint reads as
+      // localhost:11434 or `model.provider === "ollama"` exactly, which
+      // doesn't match the standardised `inference.local` URL NemoClaw routes
+      // through. Pin the compat flag here so the TUI token counter updates
+      // after each turn instead of staying `?` (#2747). Mirrors the LM Studio
+      // extension's `withLmstudioUsageCompat` workaround.
+      providerKey = MANAGED_PROVIDER_ID;
+      primaryModelRef = `${MANAGED_PROVIDER_ID}/${model}`;
+      inferenceCompat = {
+        supportsUsageInStreaming: true,
+      };
+      break;
     case "nvidia-prod":
     case "nvidia-nim":
     default: