System Architecture
Behind the scenes of Tabby.
Tabby is built with a distributed architecture to ensure low-latency interactions and powerful AI capabilities.
The Three Pillars
1. Electron Frontend
The desktop application manages global hotkeys, captures the screen and clipboard, and provides the overlay user interface. It is written in TypeScript and uses Next.js for the UI components.
2. Next.js API Backend
Acts as the central orchestrator, handling:
- Authentication via Supabase.
- AI Provider routing (OpenAI, Groq, Cerebras).
- Serving the documentation and landing pages.
3. Python Memory Backend
A specialized service using FastAPI and Mem0 to manage the persistent memory layer. It interacts with:
- Supabase (Vector Store): For fast semantic search of memories.
- Neo4j (Knowledge Graph): For mapping relationships between different concepts and preferences.
Communication Flow
- Input: A global shortcut (e.g.,
Alt+X) is pressed. - Context Capture: The Electron app captures the screen/text.
- Processing: Context is sent to the Next.js API.
- Memory Retrieval: The API queries the Memory Backend for relevant user facts.
- AI Generation: The prompt + context + memories are sent to the AI provider.
- Delivery: The response is streamed back to the Electron UI or typed out into the active app.