fix(openai-completions): enable local streaming usage compat (#68711) (thanks @gaineyllc)

src/agents/openai-completions-compat.test.ts

@@ -1,5 +1,8 @@
 import { describe, expect, it } from "vitest";
-import { resolveOpenAICompletionsCompatDefaults } from "./openai-completions-compat.js";
+import {
+  detectOpenAICompletionsCompat,
+  resolveOpenAICompletionsCompatDefaults,
+} from "./openai-completions-compat.js";
 
 describe("resolveOpenAICompletionsCompatDefaults", () => {
   it("keeps streaming usage enabled for provider-declared compatible endpoints", () => {
@@ -33,4 +36,49 @@ describe("resolveOpenAICompletionsCompatDefaults", () => {
       }).supportsUsageInStreaming,
     ).toBe(false);
   });
+
+  it.each([
+    "vllm",
+    "localai",
+    "sglang",
+    "llama-cpp",
+    "llama.cpp",
+    "llamacpp",
+    "jan",
+    "lmstudio",
+    "lm-studio",
+    "text-generation-webui",
+    "tabby",
+    "tabbyapi",
+  ])("enables streaming usage compat for known local provider %s", (provider) => {
+    expect(
+      resolveOpenAICompletionsCompatDefaults({
+        provider,
+        endpointClass: "custom",
+        knownProviderFamily: provider,
+      }).supportsUsageInStreaming,
+    ).toBe(true);
+  });
+
+  it("matches known local providers case-insensitively", () => {
+    expect(
+      resolveOpenAICompletionsCompatDefaults({
+        provider: "vLLM",
+        endpointClass: "local",
+        knownProviderFamily: "vllm",
+      }).supportsUsageInStreaming,
+    ).toBe(true);
+  });
+});
+
+describe("detectOpenAICompletionsCompat", () => {
+  it("enables streaming usage compat for vLLM on a local OpenAI-compatible endpoint", () => {
+    const detected = detectOpenAICompletionsCompat({
+      provider: "vllm",
+      baseUrl: "http://127.0.0.1:8000/v1",
+      id: "Qwen/Qwen3-Coder-Next-FP8",
+    });
+
+    expect(detected.defaults.supportsUsageInStreaming).toBe(true);
+  });
 });

src/agents/openai-completions-compat.ts

@@ -30,6 +30,27 @@ function isDefaultRouteProvider(provider: string | undefined, ...ids: string[])
   return provider !== undefined && ids.includes(provider);
 }
 
+const KNOWN_LOCAL_STREAMING_USAGE_PROVIDERS = new Set([
+  "jan",
+  "llama-cpp",
+  "llama.cpp",
+  "llamacpp",
+  "lm-studio",
+  "lmstudio",
+  "localai",
+  "sglang",
+  "tabby",
+  "tabbyapi",
+  "text-generation-webui",
+  "vllm",
+]);
+
+function isKnownLocalStreamingUsageProvider(...ids: Array<string | undefined>): boolean {
+  return ids.some(
+    (id) => id !== undefined && KNOWN_LOCAL_STREAMING_USAGE_PROVIDERS.has(id.toLowerCase()),
+  );
+}
+
 export function resolveOpenAICompletionsCompatDefaults(
   input: OpenAICompletionsCompatDefaultsInput,
 ): OpenAICompletionsCompatDefaults {
@@ -67,6 +88,10 @@ export function resolveOpenAICompletionsCompatDefaults(
     endpointClass === "mistral-public" ||
     knownProviderFamily === "mistral" ||
     (isDefaultRoute && isDefaultRouteProvider(provider, "chutes"));
+  const supportsKnownLocalStreamingUsage = isKnownLocalStreamingUsageProvider(
+    provider,
+    knownProviderFamily,
+  );
   return {
     supportsStore:
       !isNonStandard && knownProviderFamily !== "mistral" && !usesExplicitProxyLikeEndpoint,
@@ -77,7 +102,8 @@ export function resolveOpenAICompletionsCompatDefaults(
       endpointClass !== "xai-native" &&
       !usesExplicitProxyLikeEndpoint,
     supportsUsageInStreaming:
-      !isNonStandard && (!usesConfiguredNonOpenAIEndpoint || supportsNativeStreamingUsageCompat),
+      supportsKnownLocalStreamingUsage ||
+      (!isNonStandard && (!usesConfiguredNonOpenAIEndpoint || supportsNativeStreamingUsageCompat)),
     maxTokensField: usesMaxTokens ? "max_tokens" : "max_completion_tokens",
     thinkingFormat: isZai ? "zai" : isOpenRouterLike ? "openrouter" : "openai",
     visibleReasoningDetailTypes: isOpenRouterLike ? ["response.output_text", "response.text"] : [],

src/agents/openai-transport-stream.test.ts

@@ -1353,13 +1353,13 @@ 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", () => {
+  it("always includes stream_options.include_usage for known local backends like llama-cpp", () => {
     const params = buildOpenAICompletionsParams(
       {
         id: "llama-3",
         name: "Llama 3",
         api: "openai-completions",
-        provider: "custom-cpa",
+        provider: "llama-cpp",
         baseUrl: "http://localhost:8080/v1",
         reasoning: false,
         input: ["text"],