Custom AI Code Review Checks, on Open Source Models

Pricing is usage-based. You pay only for what you use, with no per-seat fees. Open source models are priced to be cheaper than frontier models for many checks, so at scale you can cut inference cost by routing routine checks to a cheaper model. Every workspace also gets $100 in free usage to start. See the docs for per-model pricing.

Yes. Each Check Run Agent is a markdown file with a model field in its frontmatter. Set kimi-k2-7-code on one agent and claude-opus-4-6 on another. Run a fast, cheap model for style and lint checks, and a deeper model for security audits. You tune cost, latency, and performance per agent.

The open source ecosystem has hit an inflection point. Top OSS models are now competitive with frontier models across many coding and reasoning tasks. Each has its own tradeoffs around cost, latency, context window, and performance. If an open source model delivers the same outcome for a given agent, you should be able to make that tradeoff yourself. Keep frontier models on the checks that need them.

They run as real GitHub check runs on every pull request. You define each check in plain English as a markdown file in .macroscope/check-run-agents/, commit it to your default branch, and it starts running on the next PR. No configuration beyond the markdown file is needed.

Connect your GitHub repository, add a markdown check definition with the model you want, and commit it. You start with $100 in free usage. It takes under 5 minutes to set up your first Check Run Agent.