Skip to content

Architecture Overview

Stimm follows a modular monolith architecture, containerized with Docker Compose. This page provides a high-level overview of the system components and their interactions.

System Diagram

graph TD
    Client[Web Client / Next.js] -->|HTTPS/WSS| Traefik[Traefik Reverse Proxy]
    Traefik -->|/api| Backend[Stimm Backend / FastAPI]
    Traefik -->|/| Frontend[Frontend Service]

    Backend --> Postgres[(PostgreSQL)]
    Backend --> Qdrant[(Qdrant Vector DB)]

    subgraph "AI Services"
        Backend -->|External API| LLM[LLM Providers]
        Backend -->|External API| TTS[TTS Providers]
        Backend -->|External API| STT[STT Providers]
    end

Core Components

1. Backend (FastAPI)

The backend is a Python FastAPI application that provides:

  • REST API for managing agents, RAG configurations, and system settings.
  • WebSocket endpoints for real-time audio streaming and conversation state.
  • Integration with LiveKit for WebRTC media transport.
  • Orchestration of AI providers (LLM, TTS, STT) and RAG retrieval.

2. Frontend (Next.js)

A modern React-based web application that offers:

  • Dashboard for managing agents and RAG configurations.
  • Real-time voice interface for conversing with agents.
  • Admin panels for system configuration.

3. Database Layer

  • PostgreSQL – Stores persistent data (agents, users, RAG configurations, documents).
  • Qdrant – Vector database for semantic search and RAG.
  • Redis – Used for caching, session storage, and SIP bridge state.

4. AI Providers

Stimm supports multiple providers for each AI task, allowing you to mix and match based on cost, latency, and quality requirements.

  • LLM: Groq, Mistral, OpenRouter, local Llama.cpp, OpenAI-compatible.
  • TTS: Deepgram, ElevenLabs, Async.ai, local Kokoro.
  • STT: Deepgram, local Whisper.

5. Real-time Media (LiveKit)

LiveKit handles WebRTC signaling, media transport, and room management. It enables low-latency, bidirectional audio streaming between clients and the backend.

6. SIP Bridge

A telephony integration that connects incoming SIP calls to LiveKit rooms, allowing phone users to interact with AI agents.

Data Flow

For a detailed step-by-step description of the audio-to-audio pipeline, see Data Flow.

Deployment

The system is designed to be deployed as a set of Docker containers orchestrated by Docker Compose. A Traefik reverse proxy routes external traffic to the appropriate services.

For production deployments, you can replace Traefik with your own ingress controller and scale services independently.

Next Steps

  • Data Flow – Understand the audio processing pipeline.
  • Components – Dive deeper into each service and module.