fix(api): repair voice notes and Veo requests

packages/api/src/providers/gemini/videogen.test.ts

@@ -73,7 +73,7 @@ describe('GeminiVideoGenProvider', () => {
     expect(request.prompt).toContain('1. hero frame');
   });
 
-  it('preserves generateAudio for text-to-video requests', async () => {
+  it('omits generateAudio for text-to-video requests because current Veo 3.1 API rejects it', async () => {
     generateVideoCalls.length = 0;
     const provider = new GeminiVideoGenProvider({
       getSecret: async () => 'test-gemini-key',
@@ -84,10 +84,10 @@ describe('GeminiVideoGenProvider', () => {
 
     expect(generateVideoCalls).toHaveLength(1);
     const request = generateVideoCalls[0] as { config?: Record<string, unknown> };
-    expect(request.config?.generateAudio).toBe(true);
+    expect(request.config?.generateAudio).toBeUndefined();
   });
 
-  it('can disable generated audio for text-to-video requests', async () => {
+  it('treats --no-audio as a compatibility no-op while generateAudio is unsupported', async () => {
     generateVideoCalls.length = 0;
     const provider = new GeminiVideoGenProvider({
       getSecret: async () => 'test-gemini-key',
@@ -98,6 +98,6 @@ describe('GeminiVideoGenProvider', () => {
 
     expect(generateVideoCalls).toHaveLength(1);
     const request = generateVideoCalls[0] as { config?: Record<string, unknown> };
-    expect(request.config?.generateAudio).toBe(false);
+    expect(request.config?.generateAudio).toBeUndefined();
   });
 });

packages/api/src/providers/gemini/videogen.ts

@@ -70,7 +70,9 @@ export class GeminiVideoGenProvider implements IVideoGenProvider {
         ...(options?.durationSec !== undefined ? { durationSeconds: options.durationSec } : {}),
         ...(options?.seed !== undefined ? { seed: options.seed } : {}),
         ...(options?.resolution !== undefined ? { resolution: options.resolution } : {}),
-        ...(!options?.imageBase64 ? { generateAudio: options?.audio !== false } : {}),
+        // The current Gemini Veo 3.1 API rejects generateAudio in this SDK path.
+        // Keep --no-audio as a no-op compatibility flag until the provider exposes
+        // a supported audio toggle again.
       },
     };
     if (options?.imageBase64) {

packages/api/src/routes/v2/__tests__/messages-send-media.test.ts

@@ -67,4 +67,44 @@ describe('POST /messages/send/media', () => {
       },
     });
   });
+
+  test('forwards WhatsApp voice-note audio as audioBuffer instead of base64', async () => {
+    const sendMessage = mock(async (_instanceId: string, _message: unknown) => ({
+      success: true,
+      messageId: 'VOICE-MSG-ID',
+      timestamp: 123,
+    }));
+    const app = mountMessagesRoutes(sendMessage);
+    const audio = Buffer.from('ogg-opus-bytes');
+
+    const res = await app.request('/messages/send/media', {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      body: JSON.stringify({
+        instanceId: '11111111-1111-4111-8111-111111111111',
+        to: '5511999999999@s.whatsapp.net',
+        type: 'audio',
+        base64: audio.toString('base64'),
+        filename: 'voice.ogg',
+        voiceNote: true,
+      }),
+    });
+
+    expect(res.status).toBe(201);
+    expect(sendMessage).toHaveBeenCalledTimes(1);
+    const message = sendMessage.mock.calls[0]?.[1] as { metadata?: Record<string, unknown> };
+    expect(message).toMatchObject({
+      content: {
+        type: 'audio',
+        filename: 'voice.ogg',
+        mimeType: 'audio/ogg; codecs=opus',
+      },
+      metadata: {
+        ptt: true,
+      },
+    });
+    expect(message.metadata?.base64).toBeUndefined();
+    expect(Buffer.isBuffer(message.metadata?.audioBuffer)).toBe(true);
+    expect((message.metadata?.audioBuffer as Buffer).equals(audio)).toBe(true);
+  });
 });

packages/api/src/routes/v2/messages.ts

@@ -533,6 +533,21 @@ const sendMediaSchema = z.object({
   threadId: z.string().optional().describe('Thread/topic ID (e.g. Telegram forum topic)'),
 });
 
+function normalizeSendMediaMimeType(data: z.infer<typeof sendMediaSchema>): string {
+  const inferred = data.mimeType ?? inferMediaMimeType(data.type, data.filename);
+  if (data.type === 'audio' && data.voiceNote === true && inferred === 'audio/ogg') {
+    return 'audio/ogg; codecs=opus';
+  }
+  return inferred;
+}
+
+function buildSendMediaMetadata(data: z.infer<typeof sendMediaSchema>): Record<string, unknown> {
+  if (data.type === 'audio' && data.voiceNote === true && data.base64) {
+    return { audioBuffer: Buffer.from(data.base64, 'base64'), ptt: true };
+  }
+  return { base64: data.base64, ptt: data.voiceNote };
+}
+
 // Send reaction schema
 const sendReactionSchema = z.object({
   instanceId: z.string().uuid().describe('Instance ID'),
@@ -1147,7 +1162,7 @@ messagesRoutes.post('/send/media', zValidator('json', sendMediaSchema), async (c
   // Resolve recipient (handles person ID to platform ID resolution)
   const resolvedTo = await resolveRecipient(data.to, instance.channel, services);
 
-  const mediaMimeType = data.mimeType ?? inferMediaMimeType(data.type, data.filename);
+  const mediaMimeType = normalizeSendMediaMimeType(data);
 
   // Build outgoing message
   const outgoingMessage: OutgoingMessage = {
@@ -1160,11 +1175,7 @@ messagesRoutes.post('/send/media', zValidator('json', sendMediaSchema), async (c
       filename: data.filename,
       mimeType: mediaMimeType,
     } as OutgoingContent,
-    metadata: {
-      base64: data.base64,
-      // WhatsApp uses 'ptt' (push-to-talk) flag for voice notes
-      ptt: data.voiceNote,
-    },
+    metadata: buildSendMediaMetadata(data),
   };
 
   // T8: API processed the send request