Onyx: Connect Team Knowledge to AI Agents

Overview

Onyx is a knowledge and context layer for teams that want AI chat and agents to use information from their organization’s apps, not just public data. It indexes connected sources with their metadata and permissions, then retrieves relevant documents for questions and research. The project is built around retrieval-augmented generation (RAG) and supports major hosted and self-hosted LLM providers.

It is aimed at teams that need a shared AI interface over internal knowledge, with deployment options designed to keep indexing and processing within their own environment. Users can access it through the web and desktop app, chat integrations, an MCP server, a Chrome extension, or an embeddable widget.

Key Features

• Index and search knowledge from more than 50 application connectors, with permissions and metadata.
• Answer questions using hybrid search and an agent harness tuned for information retrieval.
• Run multi-step deep research and augment internal sources with web search.
• Create agents with selected knowledge, custom instructions, and external actions.
• Connect to hosted or self-hosted LLM providers, including OpenAI, Anthropic, Gemini, Ollama, LiteLLM, and vLLM.
• Use a secure sandbox for code execution and generate downloadable artifacts.
• Deploy a full version for indexing and background sync, or Lite for chat and agents without document indexing.

Use Cases

• A support or operations team can search internal documents and connected apps when answering questions across dispersed knowledge sources.
• An organization can provide staff with AI agents that use selected internal knowledge and follow source permissions.
• Researchers can combine company context with web search to produce multi-step research reports.
• Developers can connect coding agents through Onyx’s MCP server to retrieve organizational context under the user’s access controls.

Getting Started

For a quick deployment, the README provides this installer command:

curl -fsSL https://onyx.app/install_onyx.sh | bash

See the README and deployment documentation for setup details and deployment options.

License in Plain Words

No standard license was detected. Without a license, the default copyright rules apply: you may read the code, but not reuse it. Check the repository or ask the authors.

Alternatives

• rag-web-ui: Focuses on document-based question answering, rather than indexing knowledge across connected apps with source permissions.
• 53AIHub: Acts as a portal for publishing and operating agents and knowledge bases, rather than a context layer that indexes connected apps.
• ragtime: Provides an OpenAI-compatible RAG API and MCP server for local knowledge, rather than a managed context layer for connected apps.

Considerations

• The project is active, with regular commits and releases, and a broad contributor base that is not concentrated in one person. Recent pull requests are being merged, while the supplied health data shows no issues closed in the measured period.
• The full deployment includes indexing, background workers, inference services, and supporting components such as Redis and MinIO, so it has more operational scope than a chat UI. Lite is described as using under 1 GB of memory but cannot index documents.
• The stated runtime requirements include Node.js and Python 3.13 or later. Check deployment documentation for the complete environment requirements of the mode you choose.
• The README describes the Community Edition as MIT-licensed and notes that Enterprise Edition adds features for larger organizations. Confirm which edition covers the features you need.
• Onyx depends on configured LLM and embedding providers; these can be hosted services or self-hosted models, so provider selection affects deployment and data-flow choices.