Metabase is not just shipping BI anymore. It is positioning analytics as something humans and agents both use, and the public pages back that up. The homepage now says “Open source analytics that answers back,” with “AI-powered analytics built on your metrics and permissions,” while the product nav still keeps the classic BI stack in view, dashboards, SQL, permissions, embeds, and cloud.
That mix matters. Metabase is not trying to replace its core business intelligence pitch. It is layering agent access on top of it. The recent AI hackathon page says the product now ships an MCP server, an Agent API, and Metabot in Slack. That is a clear bet that analytics should live inside the tools people already use, including editors and chat.
The engineering blog pushes the same story from the inside. Metabase says its backend is about 500K lines of Clojure, spread across query processing, permissions, drivers, notifications, serialization, search, and more. The team also says Claude Code alone hits the context window fast, which is why it built ten custom subagents. That is dogfooding, not a marketing demo.
The release trail supports the same read. ToolVitals shows 30 GitHub releases in 90 days and 17 release events in 30 days, with a health score of 100 and a shipping score of 100. The most recent releases in the payload, 58.13, 59.7, 59.8, 60.1, and 60.2, landed across April 20 to April 22. That is steady output, not a one-off AI stunt.
What the data does not tell you
ToolVitals can see release cadence, stars, and public positioning. It cannot see whether the AI features are actually useful, whether customers adopt them, whether the product is easy to run, or whether the semantic layer holds up under real workloads. It also cannot tell you if the new agent features are better than a polished dashboard and a good SQL editor. Publicly, Metabase is betting that the answer is yes. The data does not prove it.
How this compares to nearby tools
In the related_tools slice, Jitsu is louder on recent release activity, with 36 release events in 30 days, but much smaller on stars at 4,699. Eidos sits at the same 17 release events in 30 days as Metabase, but has 3,104 stars. Metabase is the heavyweight here, with 47,077 GitHub stars and a much broader public footprint.
Bottom line
If your team wants open source BI and is starting to care about agent-facing analytics, Metabase deserves a fresh evaluation. Its public positioning now matches the direction of travel: dashboards for people, MCP and Slack for agents, and a release cadence that says the team is actually building it. If you only want a stable dashboard layer, the AI push is not the reason to choose it. If you want analytics that can sit inside workflows, this is one to watch closely.