roborev

roborev reviews commits in the background on your own machine and keeps findings open until someone addresses them. It's for developers and teams working with AI coding agents who want feedback before a pull request, while the agent still has the code in context. The project is open source under the MIT license and runs on macOS, Linux and Windows.

It works with existing agents, including Codex, Claude Code, Gemini, Copilot and Cursor. You can choose a different agent or model for review than for writing code, use local models through Ollama or LiteLLM, and assign backup agents when a provider fails or reaches a quota. roborev doesn't require its own hosted service. Cloud-backed agents still send requests to their providers.

Reviews form a persistent ledger with severity, file locations, prompts and agent logs. Terminal and browser interfaces show what's unresolved, and reviewers can record why they disagree with a finding. Review history stays in a local SQLite database; optional PostgreSQL sync shares it across a team.

Agent hooks return findings to active coding sessions. Automated fix workflows apply changes and commit them, while branch refinement repeats review and repair in an isolated worktree. Review panels combine multiple agents into one verdict, with security and design reviews alongside analysis of existing code. Verification checks findings against current code and removes those that no longer apply.

A daemon on a server you own can also review GitHub pull requests and GitLab merge requests, post combined feedback, and record runtime, token usage and estimated cost.