From c56079348243baae9b5a4d100addd12e9ff9c756 Mon Sep 17 00:00:00 2001
From: kagura-agent <kagura.chen28@gmail.com>
Date: Sun, 19 Apr 2026 08:59:40 +0800
Subject: [PATCH] fix: always send stream_options.include_usage when streaming
 openai-completions

Backends like llama-cpp and LM Studio require stream_options: { include_usage: true }
in the request payload to report token usage in streaming responses.
buildOpenAICompletionsParams() previously gated this behind supportsUsageInStreaming
compat detection, which excluded non-standard and custom endpoints. The OpenAI SDK
sends this unconditionally, so we now do the same.

Fixes #68707
---
 src/agents/openai-transport-stream.test.ts | 31 +++++++++++++++++++++-
 src/agents/openai-transport-stream.ts      |  4 +--
 2 files changed, 31 insertions(+), 4 deletions(-)

diff --git a/src/agents/openai-transport-stream.test.ts b/src/agents/openai-transport-stream.test.ts
index 338b65be201..3aca458c73a 100644
--- a/src/agents/openai-transport-stream.test.ts
+++ b/src/agents/openai-transport-stream.test.ts
@@ -1251,6 +1251,35 @@ describe("openai transport stream", () => {
     expect(params.stream_options).toMatchObject({ include_usage: true });
   });
 
+  it("always includes stream_options.include_usage for non-standard backends like llama-cpp", () => {
+    const params = buildOpenAICompletionsParams(
+      {
+        id: "llama-3",
+        name: "Llama 3",
+        api: "openai-completions",
+        provider: "custom-cpa",
+        baseUrl: "http://localhost:8080/v1",
+        reasoning: false,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: 8192,
+        maxTokens: 4096,
+      } satisfies Model<"openai-completions">,
+      {
+        systemPrompt: "system",
+        messages: [],
+        tools: [],
+      } as never,
+      undefined,
+    ) as {
+      stream?: boolean;
+      stream_options?: { include_usage?: boolean };
+    };
+
+    expect(params.stream).toBe(true);
+    expect(params.stream_options).toEqual({ include_usage: true });
+  });
+
   it("disables developer-role-only compat defaults for configured custom proxy completions providers", () => {
     const params = buildOpenAICompletionsParams(
       {
@@ -1289,7 +1318,7 @@ describe("openai transport stream", () => {
 
     expect(params.messages?.[0]).toMatchObject({ role: "system" });
     expect(params).not.toHaveProperty("reasoning_effort");
-    expect(params).not.toHaveProperty("stream_options");
+    expect(params.stream_options).toMatchObject({ include_usage: true });
     expect(params).not.toHaveProperty("store");
     expect(params.tools?.[0]?.function).not.toHaveProperty("strict");
   });
diff --git a/src/agents/openai-transport-stream.ts b/src/agents/openai-transport-stream.ts
index baaa25c9401..f2a68fa6db1 100644
--- a/src/agents/openai-transport-stream.ts
+++ b/src/agents/openai-transport-stream.ts
@@ -1502,10 +1502,8 @@ export function buildOpenAICompletionsParams(
       ? flattenCompletionMessagesToStringContent(messages)
       : messages,
     stream: true,
+    stream_options: { include_usage: true },
   };
-  if (compat.supportsUsageInStreaming) {
-    params.stream_options = { include_usage: true };
-  }
   if (compat.supportsStore) {
     params.store = false;
   }