Open-source analytics that answers in plain English and proves it. Ask "what should I fix?" and get a computed number, not a guess. 22 reports on one page, 79 tools your coding agent can query over MCP, plus agent observability. A CI test proves the dashboard, the API and your editor return the same number. One Go binary, MIT. Self-host free, or hosted from $49/mo.

hey PH

the reason i built this: i'd look at my analytics, see signups down 9%, and just not believe it. then go check the database myself. every tool hands you a number, none of them show you how they got there.

smolanalytics answers in plain english but never lets a model write the query. it runs actual deterministic reports. and there's a CI test that fails the build if the dashboard, the api and your editor ever disagree on the same question.

someone asked me last night what that proves if the tracking itself is wrong. he was right, it doesn't. what it does is make a wrong number debuggable. one number you can chase, three you just argue about.

other half is your coding agent can query all of it. 79 tools over mcp. ask cursor "which tool call is slowest" without leaving your editor. it also watches your agent, tool latency, error taxonomy, conversation health. your agent is a user too and nobody's measuring that.

one go binary, MIT, self host free forever. or i host it, $49, no seats no sales call.

demo needs no signup, go poke at it: https://smolanalytics-demo.fly.dev

first launch, ask me anything hard

the plain english answers are actually useful, not the usual buzzword soup. Ran the CI check thing and it caught a mismatch between my dashboard and the MCP tool result which was a nice catch.

building in public here, launching tomorrow.

honest question for anyone who ships product analytics: when your dashboard says signups are down 9%, do you believe it, or do you go check in the database?