PRobe - Free, open-source AI code review companion for GitHub PRs

Hey Product Hunt! I'm Sankalp. I built PRobe in 24 hours during a jet-lagged night in India after a 21-hour flight from San Francisco, and the reason is embarrassingly simple: I'm drowning in pull request reviews. I work at a startup where a hundred-file PR is just another Tuesday. I've tried the bug bot tools that are out there and they do help, but they help the developer, not me. They leave dozens of comments about race conditions, missing null checks, incorrect loops, and that's great because it takes some of the load off my shoulders. I don't have to dig into whether someone used a while or do-while correctly anymore. But none of that cuts short my actual task as a reviewer, which has evolved into something much harder: reading between the lines. Human reviewers will always carry more context than any bot, and that's what actually matters when you're reviewing code. In my opinion, it is extremely unreliable to ship production-grade code without a human reviewer, and I think most developers would echo that sentiment. With more and more autonomous AI coding agents flooding the market, there is a staggering amount of AI slop being written today, and someone has to review it. That someone is always, inevitably, a human being who serves as the last guardrail before code hits production. So how does a human deal with this flood of AI-generated code? You cut the diamond with a diamond. You use AI to help you review the AI slop, and that's exactly what PRobe does. You chat with the pull request, ask all the questions swirling in your head, probe into specific files and even specific lines of code. But you're not just talking to a generic LLM. PRobe automatically detects the tech stack in the diff and loads open-source skill files with real coding best practices into its context. So when you ask about a .tsx file, you're essentially chatting with a senior React developer who knows the latest patterns, not a run-of-the-mill language model working off its training data alone. Once you're clear on what needs to change, you ask PRobe to leave a review and it posts the comments for you without ever leaving the page. It's free, open source, and you bring your own API keys. There's also an X-Ray mode that lets you inspect the exact system prompt sent with every message, every skill loaded, every token of context. I built it this way because that's the kind of tool I'd want to use, and I couldn't find one that worked like that, so I built it for myself. When I get back from vacation there will be a mountain of pull requests waiting for me, and that's becoming more and more of my work. Less coding, more reviewing, because the coding part is increasingly being handled by AI. Reviewing is what remains and it's only growing. This felt like a step in the right direction for my own workflow, and I'm putting it out here because I think developers like me, who are quietly becoming more reviewers than coders, could benefit from it too. Try it at getprobe.dev.