Voice Agent¶

The CodeTether Voice Agent enables real-time voice interactions with AI agents through LiveKit integration.

Overview¶

The voice agent provides: - Real-time voice streaming via LiveKit - Session playback for reviewing conversations - Multi-model support (Claude, GPT-4, Gemini, Grok) - Voice selector for choosing different AI voice models - Playback controls (pause, resume, speed adjustment)

Installation¶

Via Helm Chart¶

helm install codetether-voice-agent oci://registry.quantum-forge.net/library/codetether-voice-agent \
  --namespace codetether --create-namespace

From Source¶

cd codetether_voice_agent
pip install -r requirements.txt

Configuration¶

Environment variables:

Variable	Description	Default
`LIVEKIT_URL`	LiveKit server URL	`wss://livekit.codetether.run`
`LIVEKIT_API_KEY`	LiveKit API key	-
`LIVEKIT_API_SECRET`	LiveKit API secret	-
`VOICE_MODEL`	Default AI voice model	`claude-opus`

Usage¶

Web UI Integration¶

The voice agent integrates with the CodeTether dashboard:

import { VoiceAssistantUI } from '@/components/voice/VoiceAssistantUI';

<VoiceAssistantUI sessionId={session.id} />

Swift iOS/macOS App¶

import A2AMonitor

let voiceManager = VoiceSessionManager()
await voiceManager.connect(sessionId: "session-123")

Supported Voice Models¶

Claude (Opus, Sonnet, Haiku)
GPT-4, GPT-4 Turbo
Gemini Pro, Gemini Flash
Grok 3

Session Playback¶

Review past voice conversations:

GET /v1/voice/sessions/{id}/playback

Returns: - Audio recordings - Transcript - Agent responses - Timestamps

Architecture¶

Client (Web/Mobile)
    ↓ (WebSocket)
LiveKit Server
    ↓
Voice Agent (Python)
    ↓ (MCP)
A2A Server → Agent LLM

API Endpoints¶

Endpoint	Method	Description
`/v1/voice/sessions`	POST	Create voice session
`/v1/voice/sessions/{id}`	GET	Get session details
`/v1/voice/sessions/{id}/playback`	GET	Get playback data
`/v1/voice/models`	GET	List available models

Fine-Tuning¶

The voice agent includes fine-tuning capabilities:

cd codetether_voice_agent/finetuning
pip install -r requirements.txt
python finetune.py --data training_examples.jsonl

See finetuning/README.md for details.

Troubleshooting¶

Audio not playing?¶

Check LiveKit connectivity:

curl https://livekit.codetether.run/health

No voice response?¶

Verify model credentials:

cat ~/.local/share/agent/auth.json

Session sync issues?¶

Restart the voice agent:

kubectl rollout restart deployment/codetether-voice-agent