fix(ollama): preserve streaming usage compat

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

@@ -261,15 +261,17 @@ describe("ollama provider models", () => {
     expect(visionModel.input).toEqual(["text", "image"]);
     expect(visionModel.reasoning).toBe(true);
     expect(visionModel.compat?.supportsTools).toBe(true);
+    expect(visionModel.compat?.supportsUsageInStreaming).toBe(true);
 
     const textModel = buildOllamaModelDefinition("glm-5.1:cloud", 202752, ["completion", "tools"]);
     expect(textModel.input).toEqual(["text"]);
     expect(textModel.reasoning).toBe(false);
     expect(textModel.compat?.supportsTools).toBe(true);
+    expect(textModel.compat?.supportsUsageInStreaming).toBe(true);
 
     const noCapabilities = buildOllamaModelDefinition("unknown-model", 65536);
     expect(noCapabilities.input).toEqual(["text"]);
-    expect(noCapabilities.compat).toBeUndefined();
+    expect(noCapabilities.compat?.supportsUsageInStreaming).toBe(true);
   });
 
   it("disables tool support when Ollama capabilities omit tools", () => {
@@ -277,6 +279,7 @@ describe("ollama provider models", () => {
 
     expect(model.reasoning).toBe(false);
     expect(model.compat?.supportsTools).toBe(false);
+    expect(model.compat?.supportsUsageInStreaming).toBe(true);
   });
 
   it("parses the last positive Modelfile num_ctx value", () => {

extensions/ollama/src/provider-models.ts

@@ -249,9 +249,10 @@ export function buildOllamaModelDefinition(
       : capabilities.includes("thinking");
   const compat =
     capabilities === undefined
-      ? undefined
+      ? { supportsUsageInStreaming: true }
       : {
           supportsTools: capabilities.includes("tools"),
+          supportsUsageInStreaming: true,
         };
   return {
     id: modelId,
@@ -261,7 +262,7 @@ export function buildOllamaModelDefinition(
     cost: OLLAMA_DEFAULT_COST,
     contextWindow: contextWindow ?? OLLAMA_DEFAULT_CONTEXT_WINDOW,
     maxTokens: OLLAMA_DEFAULT_MAX_TOKENS,
-    ...(compat ? { compat } : {}),
+    compat,
   };
 }