Skip to main content
Obss shows you how well AI agents, such as Claude Code, Codex, and Cursor, do real work with your product, and which tools help them. You write tasks about your product. Obss runs each one several times, across agents, with and without your tools, and shows pass rates, cost, where agents get stuck, and a full record of every run. The repeats show whether a result holds up.

How it works

You write tasks, choose the tools to compare, and run an experiment. Then you read pass rates and cost on Matrix, check which differences are real on Comparisons, see where agents got stuck on Frictions, fix what you find, and run again. How Obss works explains each step. Here is a finished Matrix from the Stripe interface evals template. Each row is an agent and treatment, each column is a task, and each cell is a pass rate.
Matrix with the Evals metric: rows are agent and treatment combinations, columns are five tasks, and each cell shows a pass rate.Matrix with the Evals metric: rows are agent and treatment combinations, columns are five tasks, and each cell shows a pass rate.

What you can do

Ways to work with Obss

All four work on the same projects, so work you start in one shows up in the others. Secrets, such as API keys and OAuth sign-ins, stay in the dashboard. When the CLI, the plugin, or Ask Obss needs one, it gives you a link to the dashboard page where you enter it.
Obss bills in runs. Every organization starts on a free plan that resets its runs each month, and no credit card is required. Your model providers bill model use separately, through your own keys or subscriptions. See Pricing for plans, and subscribe or top up from Organization settings → Usage & billing.

Start here