Framer AI AgentsDesign and publish professional sites with AI
ReviewPhin
Self-hosted, agentic code review for GitLab and GitHub. 🐬
11 followers
Self-hosted, agentic code review for GitLab and GitHub. 🐬
11 followers
Self-hosted, agentic code review for GitLab and GitHub. Run it on your own infrastructure, bring your own model (or GitHub Copilot License), and pay per review - not per developer. A multi-agent pipeline posts findings as native, bot-owned review comments you fully control.
Hi Product Hunt!
I built ReviewPhin - self-hosted AI code review for GitLab and GitHub.
Why I built it? Tools like CodeRabbit, Greptile, and GitHub Copilot code review are genuinely great - ReviewPhin isn't trying to beat them, it fills a gap they didn't cover for me:
Per-review, not per-developer. Per-seat pricing scales with headcount even when only a fraction of MRs need AI review. One Copilot seat (or a self-hosted model) can drive reviews for a whole team.
Bring your own model - including private ones. Point the reviewer at Copilot, Azure OpenAI, Anthropic, vLLM, or a model on your own GPUs. Code stays inside your network.
GitLab-first. Most self-hosted GitLab teams get skipped. ReviewPhin starts there, with GitHub supported via a GitHub App.
How it works. You "@reviewphin review" a merge/pull request. A multi-agent pipeline (context-analyst + review-author, plus a lightweight chatter for follow-ups) produces findings with severity and inline suggestions, and comments as code reviewer. Reply in a thread to continue the conversation, or teach it durable project conventions it remembers next time.
Everything is yours - storage (SQLite or Flotiq), model, subscription, hosting. Deploy in minutes with Docker Compose or the Helm chart.
It's open source (MPL 2.0). I'd love your feedback - and if one of the hosted tools fits your team better, genuinely use them. 🐬
Love the pay-per-review model since it finally makes sense for smaller teams. One thing that would really help us is letting us customize the severity threshold or review strictness per repo, maybe through a simple config file. Some of our internal repos just need basic lint-style feedback, while others need deep architectural review, and right now a one-size approach won't cut it.
@erdi0xqk I'm glad you like the idea and thank you for your feedback!
Just to clarify, pay-per-review model is not on my end - I don't get any money from this. You are paying simply for your model usage based on how you configured the profile. So if you have already existing and underutilized license for copilot, you can get it to review GitLab merge requests without extra cost. The same if you have some deals with api usage for models on Azure or OpenAI or Anthropic.
There is also another thing worth mentioning when it comes to customization:
Reviewer will always load repo instructions, so if you have specific review instructions on how to assess severity in AGENTS.md, ReviewPhin will take that into consideration. There is also built-in memory feature. So e.g. if you write comment like:
ReviewPhin will take that, and commit to internal, per-project memory. In GitLab that memory lives as project wiki page, so you can even update it from GitLab UI later.
Somewhere in long todo list and plans I do have an idea to add support for .reviewphin directory or something similar that will allow you to add such instructions directly in the repo, but that is not there yet.