Documentation

Build a private benchmark from pull requests your team already merged, then compare your current coding-agent setup with a new one.

Install the Fullbeam GitHub App, choose repositories, and see what Fullbeam reads and writes.

Turn merged pull requests into benchmark tasks, approve specs that don't leak the answer, and validate each task before you use it.

Describe the current and new setup for each comparison flow, see which agents are supported, and add your own provider key.

Start a benchmark run for the current and new setup, follow its progress, and read pass rates, regressions and cost.

Who can see what, where benchmark runs happen, how provider keys are handled, and what Fullbeam stores.

Fix common problems with GitHub, task import, validation and comparison runs, and find answers to common questions.

Read a large pull request as chapters with cited code. A secondary, alpha feature that is separate from benchmarks.

Build a benchmark from a public GitHub repository you don't own, run agents and models on it with your own keys, and publish the results.