Simple, transparent
pricing.

Each PR targets one specific improvement — adding React.memo to a frequently re-rendered component, fixing an accessibility issue, removing an unused export, or applying an ESLint auto-fix. Changes are small, focused, and easy to review. You always review and merge manually.

Every change goes through output validation (syntax parsing) before a PR is opened. compoundbot never pushes directly to your main branch — it always creates a separate branch and PR. Your existing CI/CD runs on the PR just like any human-authored change.

Dependabot updates your dependencies. compoundbot improves your actual code — performance patterns, accessibility, security hardening, unused code removal, and framework best practices. They're complementary.

Not to get started. Install the GitHub App, pick your repos, and compoundbot runs a full audit immediately. You can optionally add a .compoundbot.yml to fine-tune which runners to use, auto-approval rules, and scan cadence.

compoundbot uses Claude Haiku for fast, cheap audit scans (discovery) and Claude Sonnet for execution (producing the actual code changes). Many runners like ESLint, Knip, and codemods are fully deterministic and don't use LLMs at all.

Yes. On the Free plan, you bring your own OpenRouter API key for public repos. Your key is encrypted at rest and only used for LLM calls on your repos. On Pro and Enterprise, managed keys are included.