Talk to your project — ask questions, give instructions, deploy fixes.

Voice Control

Codmir supports voice interaction through the Desktop app and programmatic TTS/STT clients. You can ask questions about your codebase, give instructions to agents, create tickets by speaking, and get alerted to incidents via voice barge-in.

Enable Voice in the Desktop App

Voice mode is built into the Codmir Desktop app. To activate it:

1. Open the Desktop app
2. Switch to Talk mode using the top toggle
3. Click the microphone button or press the configured hotkey
4. Start speaking -- your audio is processed through the Codmir server

The Desktop app uses Gemini Live for real-time voice sessions. Audio is streamed through the server (API keys never leave the server).

Common Voice Commands

Voice-to-Ticket Creation

Speaking a problem description automatically creates a structured ticket:

1. You say: "There's a bug where users can't reset their password on mobile"
2. Overseer creates a ticket with:
   • Title extracted from your description
   • Priority inferred from context (mobile auth = high)
   • Labels applied automatically (bug, auth, mobile)
3. You get a confirmation: "Created ticket AUTH-42: Password reset broken on mobile"

Voice-to-Agent Delegation

You can assign tasks to AI agents by voice:

1. You say: "Fix ticket AUTH-42"
2. Overseer's DecisionEngine evaluates whether to:
   • Execute silently -- fix it in the background, notify when done
   • Barge in -- if it is a critical incident, take over the voice session
   • Queue -- add it to the work queue for later
3. The agent investigates, writes a fix, runs tests, and opens a PR
4. You get a voice notification: "AUTH-42 fixed. PR #87 ready for review."

TTS Client Setup

For programmatic text-to-speech, use the TTS client from @codmir/ai:

ElevenLabs TTS

import { createTTSClient } from '@codmir/ai';

const tts = createTTSClient({
  provider: 'elevenlabs',
  apiKey: process.env.ELEVENLABS_API_KEY!,
  voiceId: 'rachel',
  model: 'eleven_turbo_v2_5',
});

// Speak text
await tts.speak('Deployment to production completed successfully.');

// With callbacks
await tts.speak('Build failed on staging.', {
  onStart: () => console.log('Speaking...'),
  onEnd: () => console.log('Done'),
});

Web Speech API (Browser)

import { WebSpeechTTSClient } from '@codmir/ai';

const tts = new WebSpeechTTSClient({
  lang: 'en-US',
  rate: 1.0,
  pitch: 1.0,
});

await tts.speak('Three new errors detected in the last hour.');

Available ElevenLabs Voices

import { ELEVENLABS_VOICES } from '@codmir/ai';

// Available voices: rachel, drew, clyde, paul, domi, bella, antoni, elli, josh, arnold, adam, sam
const voiceId = ELEVENLABS_VOICES.rachel;

Voice Notifications from Overseer

The Orchestrator can proactively speak to you via voice when incidents occur:

import { createOrchestrator } from '@codmir/overseer';

const orchestrator = createOrchestrator({
  overseer: {
    projectId: 'my-project',
    organizationId: 'my-org',
    apiKey: process.env.CODMIR_API_KEY!,
  },
  defaultRepoUrl: 'https://github.com/my-org/my-app',
  defaultBranch: 'main',
});

// Register your client for voice notifications
orchestrator.registerClient({
  userId: 'user-123',
  clientId: 'desktop-abc',
  type: 'desktop',
  capabilities: ['voice', 'notifications'],
});

await orchestrator.start();

When a critical incident occurs (error spike, service down), Overseer broadcasts to all online clients. The highest-priority client gets the voice session.

Priority order: desktop > web in voice mode > web > mobile.

Next Steps

• Self-Healing -- let voice alerts trigger autonomous fixes
• Connect Your Project -- set up the project that voice controls
• Error Tracking -- capture errors that trigger voice alerts