Sygen¶
AI assistant framework with multi-agent orchestration, background tasks, and persistent memory.
Sygen routes chat input to provider CLIs (Claude, Codex, Gemini), streams responses back via Telegram or Matrix, persists session state, and runs cron, heartbeat, webhook, and cleanup automation in-process. It also supports a direct WebSocket API transport with authenticated file upload/download.
Feature Highlights¶
- Multi-provider routing -- Claude Code, Codex, and Gemini CLI backends with automatic stream parsing and normalization.
- Telegram and Matrix transports -- Full bot integration with middleware, topic routing, streaming dispatch, and reaction buttons.
- Multi-agent orchestration -- Supervisor-managed sub-agents with inter-agent bus, shared knowledge sync, and independent workspaces.
- Background task delegation -- Long-running tasks execute in separate processes while the conversation continues.
- Persistent memory -- Session state, named sessions, and durable memory across restarts.
- Cron, webhooks, and heartbeat -- In-process automation observers for scheduled jobs, external triggers, and health monitoring.
- WebSocket API -- Direct ingress via
/wswith HTTP file endpoints for programmatic access. - Skill system and marketplace -- Extensible skill architecture with sync and discovery.
- Advanced RAG pipeline -- BM25 + vector hybrid search with ColBERT v2 reranking, query expansion, and LRU cache. 50+ languages, fully local.
- Sandboxed execution -- Optional Docker sandbox for safe command execution.
Quick Start¶
Install¶
Run the setup wizard¶
The wizard walks you through provider authentication, transport configuration (Telegram or Matrix), and optional Docker sandbox setup.
Start the bot¶
Quick Links¶
| Section | Description |
|---|---|
| Installation | Full install guide with requirements and options |
| Developer Quickstart | Shortest path for contributors |
| Configuration | Config schema, merge behavior, hot-reload |
| Architecture | Startup, routing, streaming, callbacks |
| System Overview | End-to-end mental model |
| Modules | Detailed module documentation |
Requirements¶
- Python 3.11+
- At least one authenticated provider CLI (Claude Code, Codex, or Gemini)
- Telegram bot token or Matrix credentials
- Docker (optional, recommended for sandboxing)