STELLA
System for Testing and Engineering LLM-based Conversational Agents
STELLA is a NestJS-based control plane for managing conversational AI sessions with LiveKit WebRTC integration and Kubernetes agent orchestration.
What is STELLA?
STELLA provides a complete infrastructure for building, deploying, and managing voice-enabled AI agents. It handles:
- Real-time Communication: WebRTC-based voice and data streaming via LiveKit
- Agent Orchestration: Automatic Kubernetes pod management for AI agents
- Session Management: Track conversations, participants, and messages
- Flexible Architecture: Support for multiple agent types with customizable pipelines
Architecture
Hover over a component to see details
Key Components
| Component | Description |
|---|---|
| LiveKit | External WebRTC media server (cloud-hosted or self-hosted) |
| Backend API | NestJS server managing sessions, agents, and LiveKit tokens |
| Frontend UI | React app for real-time voice/text communication |
| STT/TTS Services | Microservices for speech recognition and synthesis |
| Agent Pods | Python conversational AI agents, created per session |
Features
Project & Session Management
- Create and organize multiple projects
- Track sessions, agents, and activity per project
- Get aggregated statistics
Agent Orchestration
- Deploy conversational AI agents to Kubernetes
- Automatic Secret and Pod creation
- Pass plan configurations via environment variables
- Monitor agent status and logs
- Graceful shutdown and cleanup
LiveKit Integration
- JWT token generation for secure room access
- WebRTC-based real-time communication
- Data channel for text messages
Data Persistence
- PostgreSQL database via Prisma ORM
- Track participants, messages, and room events
- Timeline view combining messages and events
Next Steps
- Getting Started: Get STELLA running in minutes
- Build Your Own Agent: Build a custom conversational AI agent
- Architecture Overview: Deep dive into system design
- Agents Overview: Learn about different agent types