How to Self-Host Onyx AI Search on Your Own Server

Onyx (formerly Danswer) is an open-source AI search and chat platform that connects to your work tools and answers questions with citations. You can run it yourself with Docker Compose, point it at Slack, Google Drive, Confluence and dozens of other sources, and bring your own LLM. The self-hosted version is free; the cloud version is paid.

Self-hosting keeps your documents on your own infrastructure, which is the main reason teams pick it over a hosted chatbot. The cost is setup work and server resources — plan for at least 4 vCPU and 16 GB RAM for a small deployment.

What is Onyx?

Onyx is an enterprise search assistant that indexes content from connected apps and lets you ask natural-language questions. Answers come with links back to the source. It supports GPT-class and open models through API keys, and the whole stack ships as containers.

How do you install Onyx with Docker?

1. Install Docker and Docker Compose on a Linux host (Ubuntu 22.04 or similar).
2. Clone the Onyx repository and open the deployment/docker_compose folder.
3. Copy the sample .env and set a strong Postgres password plus your model API keys.
4. Run docker compose up -d and wait for the containers to become healthy.
5. Open http://your-server-ip:8080, create the first admin account.

First boot takes several minutes while migrations run. The web UI is on port 80/8080 depending on your config.

How do you connect your data sources?

In the admin panel, add connectors one at a time. Onyx will walk you through OAuth for Google Drive and Slack, or an API token for Confluence and Jira. Start with one or two sources so the first index build stays quick. Each connector has a sync schedule you can set to hourly or daily.

Which LLM should you use?

Onyx works with OpenAI, Anthropic, Azure OpenAI, and local models via Ollama or vLLM. For a private deployment, a local 7B–13B model keeps everything on-prem but answers with less polish. A hosted model gives better reasoning at the cost of sending snippets to a third party. Pick based on your data policy, not benchmarks alone.

What does it cost to run?

Software is free under an open license. Your real costs are the server and any LLM API usage. A modest VM runs the stack; indexing large drives needs more RAM and disk. Cloud plans remove the ops burden and add SSO, but you lose full data control.

Key Takeaways

• Onyx is an open-source AI search assistant you can self-host with Docker Compose.
• Minimum practical spec: about 4 vCPU, 16 GB RAM for a small team.
• Connects to Slack, Google Drive, Confluence, Jira and more with per-source sync schedules.
• Works with OpenAI, Anthropic or local models via Ollama/vLLM.
• Choose self-host for data control, cloud for less ops work.

Last updated October 2026. Compare options on the Onyx page.