Onyx is shipping like enterprise infrastructure, not a lightweight AI chat app. ToolVitals gives it a 100 health score and a 100 shipping score, and the repo shows 27 release events in 30 days, 30 releases in 90 days, and 28,893 GitHub stars. The product pitch on onyx.app is open-source AI chat connected to docs, apps, and people, but the release cadence says the harder problem is operating it.
The site also pushes deep research, MCP, code interpreter, web search, collaboration, analytics, hybrid-search, advanced RAG, contextual retrieval, and self-hosting. The GitHub repo, onyx-dot-app/onyx, uses the same positioning: open source AI platform, AI chat with advanced features, works with every LLM. That is a broad surface area, and Onyx is behaving like a team that expects customers to connect real systems, not just ask canned questions.
Recent releases back that up. v3.2.12 fixed SCIM routing, nginx routing for non-API backend paths, and perm-sync timing. v3.1.5 added tool call arg streaming, encryption key rotation, configurable file upload limits, and token-limit changes. This is platform plumbing, the kind that shows up when a product is getting used in production and the edge cases start biting.
What the data does not tell you
ToolVitals can see stars, release events, commits, and uptime. It cannot see code quality, model quality, user satisfaction, revenue, or whether the product actually answers better than your current stack. The website’s benchmark claims, including its comparisons against ChatGPT, Claude, and Notion AI, come from Onyx itself. Treat those as vendor claims until you reproduce the test.
Relative to peers
Compared with Jitsu, Onyx has a much larger audience, 28,893 stars versus 4,702, but a slightly slower recent pace, 27 release events in 30 days versus 37. Compared with dbt, Onyx is shipping harder, 100 shipping score versus 88, and has more stars, 28,893 versus 12,717. So the signal is not just activity, it is sustained maintenance on top of real attention.
Bottom line
If your team wants self-hosted AI chat over company knowledge and you can tolerate a fast-moving codebase, evaluate Onyx. If you want a static dependency, skip it. This project is being maintained like infrastructure, which is exactly why it matters.