Onyx is an open source enterprise search and AI chat platform you can self-host — a Glean alternative that connects your documents, apps, and people to any LLM for permission-aware chat, RAG search, and agents.

What is Onyx?

Onyx (formerly Danswer) is an open source AI platform that connects your company’s documents, apps, and people to a large language model for chat and search. You point it at your knowledge sources — wikis, drives, ticketing, chat — and ask questions in plain language; it retrieves the relevant passages and answers with citations back to the source.

It runs on your own infrastructure, works with any model (proprietary or self-hosted), and is MIT-licensed at its core.

What is Onyx best for?

Onyx is best for teams that want a private, self-hosted “ask your company anything” assistant instead of a vendor cloud that ingests all their data. It fits engineering, support, and operations teams that keep knowledge scattered across many tools and want one permission-aware place to search and chat over it — while keeping documents and queries on infrastructure they control.

What can Onyx do?

Connect 40+ sources out of the box — Google Drive, Confluence, Slack, GitHub, Notion, Jira, Salesforce, Microsoft Teams, and more, with real-time syncing.
Answer with cited RAG — hybrid keyword + vector search with contextual retrieval and LLM-based knowledge graphs, returning answers grounded in linked sources.
Respect permissions — every connector syncs document ACLs, so users only see answers drawn from documents they’re already allowed to open in the source app.
Run deep research — a multi-step agent flow that plans, searches, and produces longer in-depth reports.
Build custom agents — assistants with their own instructions, knowledge scope, and actions, plus MCP and external-tool support.
Use any LLM — proprietary providers or self-hosted models via Ollama, vLLM, and LiteLLM.
Do more than retrieval — web search, a code interpreter, image generation, voice mode, and downloadable artifacts.

Where does Onyx fall short?

The features large orgs need most are open-core, not MIT. Single sign-on (SAML/OIDC), fine-grained RBAC, query analytics, audit history, and white-labeling live in the paid Enterprise Edition — the fully free community build ships without them, so an enterprise rollout often means the commercial tier anyway.

Fewer connectors than the incumbents. Onyx covers 40-plus sources versus Glean’s 100+, so a very broad SaaS stack can hit gaps you’ll need to fill with custom or MCP integrations.

The lightweight mode drops the point of the product. The “Lite” deployment runs the chat UI in under 1GB of memory but leaves out indexing and RAG; the “Standard” mode that actually searches your docs runs a vector plus keyword index and background workers, which is a meaningfully heavier footprint.

Is Onyx free?

Yes — self-hosting the MIT-licensed Community Edition is free, and covers core chat, RAG search, connectors, and custom agents. Onyx Cloud is a managed option: its Business plan is $20 per user/month billed annually with a free trial, and Enterprise is custom-quoted, adding SSO, on-prem/region-specific deployment, white-labeling, and dedicated support.

What does Onyx replace?

Onyx is a self-hostable stand-in for enterprise AI assistants like Glean (roughly $50/user/month, cloud-only), ChatGPT Enterprise, and Microsoft 365 Copilot. It targets the same job — chat and search over your company’s knowledge — without per-seat cloud pricing or sending your documents to a single vendor. If you want a general chat front end rather than doc search, tools like Open WebUI and LibreChat cover that lighter use case.

FAQ

Is Onyx open source? Yes. The Onyx Community Edition is MIT-licensed and covers chat, RAG, connectors, and agents. Some enterprise-oriented features (SSO, RBAC, analytics, white-labeling) are in a separate Enterprise Edition rather than the MIT core.

Can I self-host Onyx for free? Yes. You can run the full community edition on your own servers at no license cost via Docker Compose or Kubernetes. You still pay for the compute you run it on and any paid LLM API calls you make.

Is Onyx a good Glean alternative? For teams that want self-hosting and data control, yes — it delivers permission-aware search and chat over your apps at a lower, transparent price. Glean still leads on connector breadth (100+ vs 40+) and a fully managed experience.

What do I need to run Onyx? A Docker or Kubernetes host and an LLM — either an API key for a provider like OpenAI or Anthropic, or a self-hosted model through Ollama, vLLM, or LiteLLM. The standard mode also runs a vector and keyword index plus background workers.