raven
Self-hosted AI assistant framework combining autonomous coding agent (RavenCode) with multi-channel communication and web dashboard. Python/TS/Go/Rust stack, PostgreSQL + Redis database, NATS messaging, RBAC security, and LLM failover architecture.
Ficha actualizada 22 sept 2026, 19:21 · upstream 22 sept 2026, 19:17 · release v0.4.7
Qué es raven
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
Raven is a comprehensive self-hosted AI system that serves as both an autonomous coding assistant and a multi-agent orchestration platform. It combines RavenCode — an agent capable of writing, refactoring, debugging, and executing code with full tool integration — with RavenFlow, a persistent gateway managing routing, streaming, and session management across 15 communication channels including Telegram, Discord, Slack, WhatsApp, Matrix, Teams, Signal, LINE, Feishu, IRC, Google Chat, GitHub, GitLab, email, and web chat. The system features RAG-based semantic search with BM25 keyword retrieval, a canvas visual workspace for code, tables, diagrams, and images, scheduled routines, passive monitors (HTTP endpoints, price checks, RSS feeds, file watching, process monitoring), and a full-stack React 19 dashboard served by the gateway.
- Developing software with autonomous agent assistance
- Multi-channel communication for distributed teams across Telegram, Discord, Slack, WhatsApp, Matrix, Teams, Signal, LINE, Feishu, IRC, Google Chat, GitHub, GitLab, email, and web chat
- Monitoring system health via HTTP endpoints, price checks, RSS feeds, file watching, and process monitoring
- Creating visual workflows with code, tables, mermaid diagrams, and images on a canvas workspace
- Running scheduled tasks and passive monitors without manual intervention
- Secure multi-user collaboration with RBAC, sandbox profiles, tool policies, and workspace isolation
- Real-time streaming of agent actions via WebSocket to web dashboard or other clients
- Integrating with LLM providers including OpenAI, Anthropic, OpenRouter, Ollama, vLLM, Azure, Groq, Bedrock, Vertex AI, and Copilot
- Dual architecture combining autonomous coding agent (RavenCode) with persistent multi-agent gateway (RavenFlow)
- 15 communication channels including Telegram, Discord, Slack, WhatsApp, Matrix, Teams, Signal, LINE, Feishu, IRC, Google Chat, GitHub, GitLab, email, and web chat
- Built-in text-to-speech (ElevenLabs/gTTS) and voice input with wake word detection for hands-free interaction
- RAG system with semantic search and BM25 keyword retrieval using local JSON vector store without external dependencies
- Full-stack React 19 dashboard served by the gateway, supporting real-time WebSocket streaming of agent actions
- Robust security model with JWT authentication, RBAC, sandbox profiles (main/non-main/code-exec/web-browsing/read-only), tool policy engine, SSRF protection, and Fernet encryption for secrets
- Single Python process architecture limits horizontal scaling compared to microservices
- RAG relies on local vector store which requires significant disk space (~10GB+ for large datasets)
- Tool orchestration can become complex with many concurrent tool calls
- Voice input/output introduces latency and potential privacy concerns with microphone access
- No native support for real-time collaboration or multi-user editing of shared documents
- RAG quality depends heavily on the quality of training data and embedding model choices
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Run as a single Python process using `raven start` to activate the gateway. For production, use Docker Compose with PostgreSQL (5432), NATS (18888 for messaging, 6333 for Redis), and optional monitoring stack. Set environment variables in `.env`: OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY, GROQ_API_KEY, VLLM_BASE_URL, DATABASE_URL, WEB_PORT=18888, WEB_SECRET_KEY=<hex-32>, RAVEN_MASTER_KEY=<secret>. Use `raven onboard` for interactive setup wizard. Enable RBAC with appropriate tool policies per user/channel/session. For production-grade availability, consider Kubernetes deployment or a load balancer (Traefik) in front of the gateway.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.