Pinrail - Desktop inbox for coding agent reviews
Show HN: Pinrail – A desktop inbox where coding agents wait for your review
Pinrail is a desktop app that puts a human in the loop for agentic workflows. Instead of approving critical steps in a chat, your coding agent submits a review to Pinrail's inbox, where you decide in a view made for the content—diffs, lists, images, or documents. It works with any agent that can run a command, such as Claude Code, Codex, Cursor, or OpenCode, and keeps everything on your machine. The structured decision returns to the agent as Markdown or JSON, so it can carry on. Available for macOS and Linux, with Windows support coming.
Pinrail gives each of these moments a proper review: the agent decides when to ask, each review has a view made for its content, your decision is structured, and everything stays on your machine.
- Jayakumark
This looks great, can this be a claude mod ?
- hensenjuang
Is there a way to mark an approval as stale if the diff changes while you're reviewing it?
- pnezis
Hi HN, author here.
Agents can now do real work on their own: they review pull requests, draft emails, write documents and generate images. The hard part is no longer getting the work done, but reviewing it well before it is used.
Pinrail is a desktop app for that review. The agent submits what it made and waits. You see it in a view made for that kind of content, such as comments on a diff, a set of draft emails or an image you can mark up, and you decide item by item. The agent then continues with your decision.
I built it because I kept approving agent work in the worst possible place: a terminal scrollback. When the agent drafted twelve review comments for a pull request, I either let it post all of them or argued about them one by one in chat. Neither worked well, so I had the agent write its comments into a custom HTML page that showed them on the diff. There I could accept some, reject others with a reason, edit one, and hand the whole decision back at once. That page worked so well that I wanted the same for everything else my agents asked me about, and that is how Pinrail started.
It works like this. In the agent's instructions you name the steps that need a person, for example "before posting review comments, submit them to Pinrail and wait". When the agent reaches that step, it runs:
pinrail submit code-review --title "Dedup tickets on save" --data proposals.json --wait
The review appears in the app's inbox and you get a notification. You decide in a view made […]