A sophisticated, multi-platform conversational AI agent with voice capabilities, multiple LLM provider support, and enterprise-grade reliability features.
- WebSocket WebUI - Modern browser-based interface with real-time chat
- Telegram Integration - Full Telegram bot support with voice messages
- Unified Session Management - Seamless user experience across platforms
- Primary Provider: OpenRouter (access to 400+ models)
- Automatic Fallback: Mistral AI with intelligent failover
- Dynamic Switching: Change models mid-conversation
- Fine-Grained Control: Temperature, system prompt, memory customization
- Text-to-Speech: ElevenLabs integration
- Speech-to-Text: Deepgram Nova-2 transcription
- Dual Platform: Voice on Telegram and WebUI
- Rate Limiting: 100 requests/minute per session
- Input Validation: Comprehensive validation of all commands
- Error Recovery: Graceful shutdown and resource cleanup
- Structured Logging: JSON logs with request tracking
/healthendpoint for system monitoring/metricsendpoint for real-time metrics- Structured JSON logging to
logs/app.logandlogs/error.log - Request latency and performance tracking
# Install
npm install
# Configure
cp .env.example .env # Add your API keys
# Run
npm startVisit http://localhost:3000
Set these environment variables in .env:
# At least one LLM provider required
OPENROUTER_API_KEY=your_key
OPENROUTER_MODEL=anthropic/claude-3.5-sonnet
MISTRAL_API_KEY=your_key
# Optional but recommended
REDIS_URL=redis://localhost:6379
TELEGRAM_BOT_TOKEN=your_token
ELEVENLABS_API_KEY=your_key
DEEPGRAM_API_KEY=your_key
# Server config
PORT=3000
NODE_ENV=production/settings- View configuration/provider <openrouter|mistral>- Switch AI provider/model <name>- Change model/temp <0-2>- Set creativity level/system <prompt>- Custom personality/save- Save your settings/help- List all commands
# Health check
curl http://localhost:3000/health
# Metrics
curl http://localhost:3000/metrics
# Logs
tail -f logs/app.logSee CLAUDE.md for detailed development guide.
MIT