Run it yourself, or let me run it.

Every edition has the whole product: the collector, KPIs as code, the MCP server your agent analyses through, prompts and evaluation. Editions differ only in who runs the server and how much data it keeps.

Self-hosted

Community

Free and open source (AGPL), on your own server

No limits: websites, events, people, data kept
Your data never leaves your infrastructure
One Rust binary: SQLite to try it, Postgres for production
Your coding agent analyses, with its own model; built-in checks without one

trace init && trace serve Get the code

Hosted

Nothing to install. Same product, on my servers.

Free

To try it on a real website

30 days of data kept
1 website
100,000 events a month
3 people
Analysis by your coding agent, unlimited

Start free

Paid

Talk to me: limits set with you, per account

Data kept for as long as you need
Several websites under one account
Higher event and analysis limits
Two-step required for your whole team

Contact

Questions

What happens when I hit the event limit?

Your website keeps working exactly as before. Trace stops recording for the rest of the month and tells you how many events it skipped.

Why is there no checkout?

Paid plans start with a conversation. Write to me and I set your limits by hand.

Can I move from hosted to self-hosted?

Yes. It is the same software. Your trace.json and snippet stay the same; only the server address changes.

Does my agent see my users?

No. Over MCP your agent reads the same aggregate summary you see on the Behaviour page: counts and rates per page, never sessions or people. It analyses with its own model, so the summary goes only where you send it.