fix(gateway): fall back to lastCallUsage on /v1/chat/completions

src/gateway/openai-http.ts

@@ -1,7 +1,12 @@
 import { randomUUID } from "node:crypto";
 import type { IncomingMessage, ServerResponse } from "node:http";
 import type { ImageContent } from "../agents/command/types.js";
-import { normalizeUsage, toOpenAiChatCompletionsUsage } from "../agents/usage.js";
+import {
+  hasNonzeroUsage,
+  normalizeUsage,
+  toOpenAiChatCompletionsUsage,
+  type NormalizedUsage,
+} from "../agents/usage.js";
 import { createDefaultDeps } from "../cli/deps.js";
 import { agentCommandFromIngress } from "../commands/agent.js";
 import type { GatewayHttpChatCompletionsConfig } from "../config/types.gateway.js";
@@ -369,6 +374,7 @@ async function resolveImagesForRequest(
 export const __testOnlyOpenAiHttp = {
   resolveImagesForRequest,
   resolveOpenAiChatCompletionsLimits,
+  resolveChatCompletionUsage,
 };
 
 function buildAgentPrompt(
@@ -466,16 +472,26 @@ type AgentUsageMeta = {
   total?: number;
 };
 
-function resolveRawAgentUsage(result: unknown): AgentUsageMeta | undefined {
-  return (
+function resolveAgentRunUsage(result: unknown): NormalizedUsage | undefined {
+  const agentMeta = (
     result as {
       meta?: {
         agentMeta?: {
           usage?: AgentUsageMeta;
+          lastCallUsage?: AgentUsageMeta;
         };
       };
     } | null
-  )?.meta?.agentMeta?.usage;
+  )?.meta?.agentMeta;
+  const primary = normalizeUsage(agentMeta?.usage);
+  if (hasNonzeroUsage(primary)) {
+    return primary;
+  }
+  const fallback = normalizeUsage(agentMeta?.lastCallUsage);
+  if (hasNonzeroUsage(fallback)) {
+    return fallback;
+  }
+  return primary ?? fallback;
 }
 
 function resolveChatCompletionUsage(result: unknown): {
@@ -483,7 +499,7 @@ function resolveChatCompletionUsage(result: unknown): {
   completion_tokens: number;
   total_tokens: number;
 } {
-  return toOpenAiChatCompletionsUsage(normalizeUsage(resolveRawAgentUsage(result)));
+  return toOpenAiChatCompletionsUsage(resolveAgentRunUsage(result));
 }
 
 function resolveIncludeUsageForStreaming(payload: OpenAiChatCompletionRequest): boolean {

src/gateway/openai-http.usage.test.ts

@@ -0,0 +1,78 @@
+import { describe, expect, it } from "vitest";
+import { __testOnlyOpenAiHttp } from "./openai-http.js";
+
+const { resolveChatCompletionUsage } = __testOnlyOpenAiHttp;
+
+describe("resolveChatCompletionUsage", () => {
+  it("maps agentMeta.usage to OpenAI prompt/completion/total fields", () => {
+    const result = {
+      meta: {
+        agentMeta: {
+          usage: { input: 120, output: 42, cacheRead: 10, total: 172 },
+        },
+      },
+    };
+
+    expect(resolveChatCompletionUsage(result)).toEqual({
+      prompt_tokens: 130,
+      completion_tokens: 42,
+      total_tokens: 172,
+    });
+  });
+
+  it("falls back to agentMeta.lastCallUsage when agentMeta.usage is missing", () => {
+    const result = {
+      meta: {
+        agentMeta: {
+          lastCallUsage: { input: 80, output: 20, total: 100 },
+        },
+      },
+    };
+
+    expect(resolveChatCompletionUsage(result)).toEqual({
+      prompt_tokens: 80,
+      completion_tokens: 20,
+      total_tokens: 100,
+    });
+  });
+
+  it("falls back to agentMeta.lastCallUsage when agentMeta.usage is all zero", () => {
+    const result = {
+      meta: {
+        agentMeta: {
+          usage: { input: 0, output: 0, total: 0 },
+          lastCallUsage: { input: 55, output: 7, total: 62 },
+        },
+      },
+    };
+
+    expect(resolveChatCompletionUsage(result)).toEqual({
+      prompt_tokens: 55,
+      completion_tokens: 7,
+      total_tokens: 62,
+    });
+  });
+
+  it("returns zeros when both agentMeta.usage and lastCallUsage are absent", () => {
+    const result = { meta: { agentMeta: {} } };
+
+    expect(resolveChatCompletionUsage(result)).toEqual({
+      prompt_tokens: 0,
+      completion_tokens: 0,
+      total_tokens: 0,
+    });
+  });
+
+  it("returns zeros when the result has no meta at all", () => {
+    expect(resolveChatCompletionUsage({})).toEqual({
+      prompt_tokens: 0,
+      completion_tokens: 0,
+      total_tokens: 0,
+    });
+    expect(resolveChatCompletionUsage(null)).toEqual({
+      prompt_tokens: 0,
+      completion_tokens: 0,
+      total_tokens: 0,
+    });
+  });
+});