The latest articles on DEV Community by Vijay Soni (@vs4vijay). https://dev.to/vs4vijay https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F143418%2Fd16ee030-50fe-4289-9eb6-1a61c17b0e8b.jpeg https://dev.to/vs4vijay en Vijay Soni Sat, 10 Oct 2026 20:30:00 +0000 https://dev.to/vs4vijay/show-dev-prx-local-first-pr-review-ai-informs-human-decides-4k7p https://dev.to/vs4vijay/show-dev-prx-local-first-pr-review-ai-informs-human-decides-4k7p <p>Pull request reviews pile up. For me the slow part was never the reading. It was the switch: opening a PR cold, working out what it was for, and finding the few things in it that mattered. By the time I had the context, I had lost the thread of whatever I was doing before.</p> <p>I wanted the first pass done before I got there. I did not want to hand an AI the keys to my repositories to get it.</p> <p>So I built <a href="https://github.com/vs4vijay/prx" rel="noopener noreferrer">prx</a>. Its tagline is "AI informs, human decides", and this post is about what that means in the code.</p> <h2> What it does </h2> <p>prx runs on your machine. A background daemon polls the GitHub repositories you choose. For each open pull request it runs a coding-agent CLI you already have installed (<code>claude</code>, <code>codex</code> or <code>pi</code>), or calls an LLM API, and gets back a review.</p> <p>When I sit down to review, each PR already has:</p> <ul> <li>a brief: a few plain sentences on what the PR changes and what to notice,</li> <li>findings, each marked <code>blocking</code> or <code>nit</code>,</li> <li>a one-line verdict.</li> </ul> <p>On macOS it sits in the menu bar. Everywhere, <code>prx status</code> and <code>prx brief</code> show the same thing in a terminal.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5kbhtnbouny1hby6ftdu.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5kbhtnbouny1hby6ftdu.png" alt="The PRX menu bar menu" width="800" height="944"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fngzcag0ql6dd46n3no6p.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fngzcag0ql6dd46n3no6p.png" alt="The PRX prs" width="800" height="552"></a></p> <h2> The problem with a reviewer that can post </h2> <p>An agent that can post to GitHub can also approve a pull request, merge it, or follow instructions hidden in a diff. I did not want to depend on the agent behaving.</p> <h2> What that does not give you </h2> <p>I want to be exact about the limits.</p> <p>The agent's read-only rule is an instruction in a prompt. An agent CLI runs as you, with the tools and permissions you start it with, and the built-in <code>claude</code> engine runs with <code>--dangerously-skip-permissions</code> so a background review never stops to ask. prx does not sandbox the agent. If you want an engine with no tools at all, use an LLM API engine: it receives the prompt and the diff and can only return text.</p> <p>The automated review is posted from your GitHub account, with a footer that says it is automated, and it can be a request-changes review. If you would rather see reviews without posting anything, there is a <code>monitor</code> mode, and <code>prx review --dry-run</code> prints what would be posted.</p> <p>"Local-first" describes prx: the daemon, its state and its history are on your disk, there is no server of mine in the middle, and there is no telemetry. GitHub is reached through the <code>gh</code> session you are already signed in to, so prx stores no GitHub token. The diff still goes to whichever model provider your agent or API uses. A local OpenAI-compatible server such as Ollama keeps that on your machine too.</p> <h2> What it runs on </h2> <p>One Rust binary holds the CLI, the daemon, the per-PR state machine and a localhost API. The macOS menu bar app is a separate, thin client of that API. It runs on macOS and Linux. There is no Windows build, and GitHub is the only forge.</p> <p>You bring the agent or the API key. prx does not include a model, so a review costs whatever yours charges.</p> <p>This is v0.1.0 and I built it alone. The Mac app is not notarized yet, so a downloaded copy needs one command before it opens; the README has it. Linux binaries need glibc 2.39 or later.</p> <p>Give it a try - <a href="https://github.com/vs4vijay/prx" rel="noopener noreferrer">https://github.com/vs4vijay/prx</a></p> showdev opensource github ai