Hey Product Hunt 👋 I'm Dharmesh, and I build CodeOtter. Why I built it: every AI reviewer I tried asked one chat model to do two jobs, write a nuanced review AND give the PR a score. The scores always landed between 75 and 85, whether the diff was a typo or a risky auth refactor, so nobody trusted them. CodeOtter splits the work: • A small deterministic engine owns the numbers: 6 gauges from 0 to 100 (Quality, Blast Radius, Risk, Tests, Readability, PR Hygiene) and 9 yes/no merge gates. • An LLM writes the review, line-anchored findings and one-click GitHub suggestions. It runs on your own hardware with Docker. Use local GGUF models (fully offline) or bring your own key (Claude, OpenAI, MiniMax, OpenRouter). It reads your AGENTS.md / CLAUDE.md as repo rules, remembers findings you dismiss, and works with GitHub, Forgejo and Gitea. No telemetry, no per-seat billing, open source under AGPL-3.0. It's young (a few weeks old), so I'd really value honest feedback, especially from teams who'd want to run AI review on-prem or air-gapped. What would stop you from trying it?

CodeOtter

Self-hosted AI code review with calibrated PR scores

Self-hosted AI code review with calibrated PR scores