From aab1435cb3d1d665de69c8d56cb5a279c0134390 Mon Sep 17 00:00:00 2001
From: Mohamed Boudra <boudra.moha@gmail.com>
Date: Mon, 8 Jun 2026 14:03:15 +0700
Subject: [PATCH] Fix OpenAI speech streaming after SDK upgrade

---
 package-lock.json                             |  2 +-
 packages/server/package.json                  |  2 +-
 .../speech/providers/openai/tts.test.ts       | 62 +++++++++++++++++++
 .../src/server/speech/providers/openai/tts.ts |  7 ++-
 4 files changed, 70 insertions(+), 3 deletions(-)
 create mode 100644 packages/server/src/server/speech/providers/openai/tts.test.ts

diff --git a/package-lock.json b/package-lock.json
index cc9d08748..c8af37911 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -38150,7 +38150,7 @@
       "version": "0.1.91-beta.2",
       "dependencies": {
         "@agentclientprotocol/sdk": "^0.17.1",
-        "@anthropic-ai/claude-agent-sdk": "0.3.168",
+        "@anthropic-ai/claude-agent-sdk": "^0.3.168",
         "@anthropic-ai/sdk": "^0.102.0",
         "@getpaseo/client": "0.1.91-beta.2",
         "@getpaseo/highlight": "0.1.91-beta.2",
diff --git a/packages/server/package.json b/packages/server/package.json
index 14836c098..724976116 100644
--- a/packages/server/package.json
+++ b/packages/server/package.json
@@ -58,7 +58,7 @@
   },
   "dependencies": {
     "@agentclientprotocol/sdk": "^0.17.1",
-    "@anthropic-ai/claude-agent-sdk": "0.3.168",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.168",
     "@anthropic-ai/sdk": "^0.102.0",
     "@getpaseo/client": "0.1.91-beta.2",
     "@getpaseo/highlight": "0.1.91-beta.2",
diff --git a/packages/server/src/server/speech/providers/openai/tts.test.ts b/packages/server/src/server/speech/providers/openai/tts.test.ts
new file mode 100644
index 000000000..7f9a335b0
--- /dev/null
+++ b/packages/server/src/server/speech/providers/openai/tts.test.ts
@@ -0,0 +1,62 @@
+import type pino from "pino";
+import { ReadableStream } from "node:stream/web";
+import { beforeEach, describe, expect, test, vi } from "vitest";
+
+const openAiMocks = vi.hoisted(() => ({
+  createSpeech: vi.fn(),
+}));
+
+vi.mock("openai", () => ({
+  OpenAI: vi.fn().mockImplementation(
+    class {
+      audio = {
+        speech: {
+          create: openAiMocks.createSpeech,
+        },
+      };
+    },
+  ),
+}));
+
+import { OpenAITTS } from "./tts.js";
+
+function createLogger(): pino.Logger {
+  const logger = {
+    child: vi.fn(),
+    debug: vi.fn(),
+    error: vi.fn(),
+    info: vi.fn(),
+  };
+
+  logger.child.mockReturnValue(logger);
+
+  return logger as unknown as pino.Logger;
+}
+
+describe("OpenAITTS", () => {
+  beforeEach(() => {
+    openAiMocks.createSpeech.mockReset();
+  });
+
+  test("returns a Node stream from the OpenAI web response body", async () => {
+    openAiMocks.createSpeech.mockResolvedValue({
+      body: new ReadableStream<Uint8Array>({
+        start(controller) {
+          controller.enqueue(Buffer.from("audio"));
+          controller.close();
+        },
+      }),
+    });
+
+    const tts = new OpenAITTS({ apiKey: "sk-test" }, createLogger());
+    const result = await tts.synthesizeSpeech("hello");
+
+    const chunks: Buffer[] = [];
+    for await (const chunk of result.stream) {
+      chunks.push(Buffer.from(chunk));
+    }
+
+    expect(Buffer.concat(chunks).toString()).toBe("audio");
+    expect(result.format).toBe("pcm");
+  });
+});
diff --git a/packages/server/src/server/speech/providers/openai/tts.ts b/packages/server/src/server/speech/providers/openai/tts.ts
index f7cc7dd39..9813b4760 100644
--- a/packages/server/src/server/speech/providers/openai/tts.ts
+++ b/packages/server/src/server/speech/providers/openai/tts.ts
@@ -1,6 +1,7 @@
 import type pino from "pino";
 import { OpenAI } from "openai";
 import { Readable } from "node:stream";
+import type { ReadableStream as NodeReadableStream } from "node:stream/web";
 import type { SpeechStreamResult, TextToSpeechProvider } from "../../speech-provider.js";
 
 export type { SpeechStreamResult };
@@ -65,7 +66,11 @@ export class OpenAITTS implements TextToSpeechProvider {
           | "pcm",
       });
 
-      const audioStream = response.body as unknown as Readable;
+      if (!response.body) {
+        throw new Error("OpenAI speech response did not include an audio stream");
+      }
+
+      const audioStream = Readable.fromWeb(response.body as NodeReadableStream<Uint8Array>);
 
       const duration = Date.now() - startTime;
       this.logger.debug({ duration }, "Speech synthesis stream ready");