Onyx (formerly Danswer, YC W24) is a comprehensive open-source AI platform with over 20,000 GitHub stars that goes far beyond a simple chat interface. It combines enterprise search, hybrid RAG with knowledge graphs, custom AI agents, deep research capabilities, MCP tool support, code interpretation, and image generation into a single self-hosted deployment. The platform connects to 40+ data sources — Slack, Confluence, Jira, Google Drive, SharePoint, Salesforce, GitHub, Notion, and many more — with real-time synchronization and permission inheritance that respects who can see what in the original source system.
The RAG pipeline is where Onyx particularly distinguishes itself. Instead of basic vector similarity search, it employs hybrid retrieval combining keyword and semantic approaches, contextual retrieval that considers document structure, and LLM-generated knowledge graphs that capture entity relationships across the indexed corpus. This multi-layered approach maintains accuracy even at scales of tens of millions of documents, making it suitable for large enterprise deployments. The agent framework lets teams build specialized AI assistants that can search internal knowledge, browse the web, execute code, interact with external systems via MCP, and chain these capabilities into complex workflows.
Deployment options range from a simple Docker Compose setup for small teams to Kubernetes with Helm charts for enterprise scale, including fully airgapped environments. The Community Edition is MIT licensed and fully functional, while the Enterprise Edition adds SSO, advanced RBAC, white-labeling, and dedicated support. Onyx Cloud offers managed hosting at $20 per user per month. For teams evaluating self-hosted AI platforms, Onyx represents one of the most complete open-source options — combining the chat UI of Open WebUI, the RAG capabilities of LangChain, and the enterprise features of commercial platforms into a single integrated system.