A Better Way to Handle Code Reviews with Plannotator
It’s not just for local work either. It handles branches, worktrees, and even pulls in GitHub or GitLab PRs. I love that the PR view lets you filter out all that annoying bot noise so you can actually focus on the human comments and descriptions.
The tool seems to be built in two layers: a rock-solid manual review tool and an optional AI layer that honestly feels like magic. You can ask the AI questions inline, or use "Guided Reviews" where an agent takes a massive, overwhelming diff and reorganizes it into a semantic overview based on what actually matters. It's almost like having a senior dev prep the context for you.
For the power users, the "Review Agents" feature is a game changer. You can run specific agents that drop findings directly as inline annotations. Plus, since there’s a local API, you can have your own agents POST insights or notes onto the diff. Everything you do is saved locally to ~/.plannotator, which means you can eventually train a "custom review skill" through what they call compound engineering. Basically, the more you use it, the smarter your local feedback loop gets.
It’s free, uses MIT/Apache 2.0 licenses, and runs entirely on your machine, so privacy isn't an issue. If you're looking to integrate AI agents into your development workflow without losing control, you should definitely check out the repo: https://github.com/backnotprop/plannotator
Has anyone else tried using local agents to help with their PR workflow yet? I'm curious to see how much time this actually saves in a real-world production environment.