Launching today
burnban
Meter and cap AI agent spend on your own machine
9 followers
Meter and cap AI agent spend on your own machine
9 followers
burnban is the local instrument for AI agent spend. One binary sits in the request path, meters every token, and caps budgets before they blow instead of after. Nothing leaves your machine, so there is no dashboard to log into and no traffic to hand over. It also prices your flat-rate plan against real metered rates, so you finally see what those tokens are worth. Open source and MIT licensed. Sync and team budgets are next; the waitlist at burnban.dev sees them first.
Builder here. Last month my machine ran $4,173.49 of API-equivalent agent usage on a $200/month plan. Nothing local could tell me that, so I built the thing that could.
Burnban is one Go binary, local, MIT.
burnban usage prices the logs nine agent CLIs already keep on disk (read-only, no account).
burnban serve meters API-key traffic through a localhost proxy and refuses to forward once you cross your caps; fuses catch loops and fan-out before they get expensive. SQLite ledger, keys never stored, dashboard is one embedded HTML file.
Everything in this release is free. Tell me where it's wrong and I'll fix it.
one thing that would honestly seal the deal for me is a per-agent or per-model budget split, so you can cap that gpt-4 spend separately from the cheaper models without writing yaml gymnastics. feels like a natural next step after the global cap.
Demo link, interactive.
https://burnban.dev/demo/
Added light mode in the last minute. :)