daydream
An automated code-review command-line agent that reviews a change, applies fixes, and runs tests, recording every step as a structured trajectory used to train an open-weight code-review model.
daydream reviews a code change, applies fixes, and runs the project's test suite to confirm the result, recording every action it takes as a structured trajectory. Those trajectories accumulate into a training corpus that the project uses to train an open-weight code-review model through a staged recipe of reward construction, supervised fine-tuning, rejection fine-tuning, and online reinforcement learning, benchmarked against commercial code-review bots on held-out pull requests.
It requires Python and the Claude Code command-line interface, with optional support for the GitHub CLI, Codex CLI, and other backends. It installs as an editable command-line tool from a cloned repository and is run against a target project directory to trigger a full review, fix, and test cycle.