Onyx

What does Onyx do?

Onyx describes itself as "the application layer for LLMs" and says it "enables LLMs through advanced capabilities like RAG, web search, code execution, file creation, deep research and more." Search over company knowledge uses "over 50+ indexing based connectors provided out of the box or via MCP" and agentic retrieval over a hybrid index with AI agents. Deep research writes multi-step reports; the README says it was top of a leaderboard in February 2026, the project's own claim, and that its search benchmark is still to be published.

Custom agents carry their own instructions, knowledge and actions, and actions and MCP reach outside applications. Web search runs through Serper, Google PSE, Brave, SearXNG and others, with its own crawler. It adds sandboxed code execution, voice mode, image generation and downloadable artifacts. It works with "all major LLM providers, both self-hosted (like Ollama, LiteLLM, vLLM, etc.) and proprietary (like Anthropic, OpenAI, Gemini, etc.)." The Enterprise Edition adds single sign-on, SCIM provisioning, role-based access, analytics, query history and whitelabeling.

Key facts

• Licence: MIT outside the ee directories; the Onyx Enterprise License inside them, which needs a paid seat licence for production use and forbids distributing that code.
• The README says the Community Edition covers all the core features for chat, RAG, agents and actions.
• 50+ connectors, out of the box or through MCP, by the README.
• Agentic retrieval over a hybrid vector and keyword index; multi-step deep research.
• Custom agents, actions and MCP; web search through Serper, Google PSE, Brave, SearXNG and others; code execution, voice and image generation.
• Models: self-hosted through Ollama, LiteLLM or vLLM, or hosted from Anthropic, OpenAI, Gemini and others.
• Enterprise Edition: SSO with Google OAuth, OIDC or SAML, SCIM, role-based access, analytics and whitelabeling. Onyx Cloud offers a free trial.
• Install: one command with the install script, or Docker, Kubernetes, Helm or Terraform; a Lite mode runs in under 1 GB of memory without the search index. Latest tagged release when read: v4.7.7 on 2026-09-16; 32,149 stars.

What does it replace, and where does it fall short?

Onyx replaces Glean for a company that wants one place to search and ask questions across its tools and documents, with agents and deep research, on its own servers and with the model of its choice, local or hosted. It is the closest open-source match on the open-source alternatives to Glean page.

Where it falls short: single sign-on, SCIM, role-based access and analytics, which most companies need for a Glean replacement, are Enterprise Edition features under a paid licence. The Lite mode leaves out the search index. The leaderboard claim is the project's own and its benchmark is unpublished. For a free option with multi-user sign-in built in, Open WebUI.

How does Onyx run?

Run the install script with one command, or deploy with Docker, Kubernetes, Helm or Terraform. The Lite mode needs under 1 GB of memory and acts as a light chat interface; the standard mode adds the vector and keyword index, background workers, model servers, Redis and MinIO. Onyx Cloud lets you try it without deploying, and the Enterprise Edition is a paid licence.

Who is Onyx for?

A company that wants Glean's search and answers across its knowledge sources, self-hosted, and either accepts the Community Edition's features or buys the Enterprise licence for single sign-on and roles. A small team that only needs chat over a few documents should look at AnythingLLM.

What limits does the README state?

From the README: Lite mode requires fewer resources and leaves out the vector and keyword index for retrieval; the search benchmark is marked "Benchmark to release soon!"

Questions people ask

Is Onyx open source?

The Community Edition is, under MIT, and the README says it covers all the core features for chat, retrieval, agents and actions. Code in the ee directories is under the Onyx Enterprise License, which needs a paid seat licence for production use. This site labels Onyx open source and states the split.

Is Onyx the same as Danswer?

Yes. Onyx was formerly called Danswer, and the danswer-ai/danswer address resolves to the same repository with the same star count. The product, the licence split between MIT and the enterprise directories, and the connectors are all the same; only the name changed.

Can Onyx use local models?

Yes. The README says it works with self-hosted models through Ollama, LiteLLM and vLLM, and with hosted ones from Anthropic, OpenAI, Gemini and others, so company data and the model can both stay on your own servers.

How does Onyx compare with Glean?

Both search and answer across a company's knowledge. Onyx does it self-hosted with 50+ connectors, agents and deep research, and any model; its single sign-on, roles and analytics need the paid Enterprise Edition. The open-source alternatives to Glean page compares it with the other options.

What does Onyx need to run?

The Lite mode needs under 1 GB of memory and works as a light chat interface. The standard mode adds a vector and keyword index, background workers, model servers, Redis and MinIO. One install script sets it up, or you deploy with Docker, Kubernetes, Helm or Terraform.

Sources

• Onyx README, LICENSE and backend/ee/LICENSE: github.com/onyx-dot-app/onyx, read 2026-09-17.