# RunThru

> RunThru is a GitHub app that verifies pull requests by using them: it films a narrated walkthrough of the change on the preview deployment, and a second model checks that the feature works. Comment `/runthru` on a pull request: an AI agent opens the preview deployment in a real browser, finds the flow the change affects, writes and voices a script, and films it with a cursor, camera zooms, captions and chapters. A video model then watches the result and posts a QA verdict ("feature works", score out of 10, timestamped issues) with the video on the pull request, in about 5 to 8 minutes.

Key facts:

- Trigger: comment `/runthru` on a GitHub pull request, optionally with direction in plain English (`/runthru show the new export menu, then download a CSV`) or a URL.
- Finds deploy previews from Vercel, Netlify, Render and Cloudflare Pages automatically; also any URL, a URL pattern, or a static folder via `.runthru.yml`.
- Output: 1080p60 MP4 (or 4K), GIF preview, chapters, word-synced captions, transcript, and 1:1 and 9:16 cuts.
- Sharing: every video has a share page (runthru.co/v/...) anyone with the link can watch, with copy-link, email and download buttons. Slack notifications post each video with its verdict to a channel (every video, or only the ones that need attention); set up at https://www.runthru.co/connect, then comment `/runthru connect <code>`.
- QA: Gemini 3.1 Pro watches the finished video next to the script, flags narration that doesn't match the screen and errors in the app itself, and writes `judge.json` (score, featureWorks, issues).
- Models: Claude Opus 5.5 explores and writes; ElevenLabs voices; Gemini reviews.
- Security: never pushes code, never runs repository code, only runs when an owner, member or collaborator comments, and blocks private-network addresses in its browser.
- Pricing is per video, not per seat: Free $0 (5 videos/month, 720p), Pro $29/month (30 videos, then $1), Team $99/month (100 videos, then $0.90), Enterprise custom (self-hosted CLI and GitHub Action, your own model keys, localhost and iOS Simulator).
- Good fit: teams shipping UI weekly, reviewers of AI-agent-written pull requests (Claude Code, Cursor, Codex, Copilot), developer-tool companies making changelog and launch videos, solo builders who want a launch video without screen recording.
- Not a fit: talking-head async video messages (use a screen recorder for those).

## Docs

- [Docs](https://www.runthru.co/docs): quickstart, comment commands, preview detection, `.runthru.yml` reference, logins and protected previews, outputs, share pages and Slack, limits, permissions, self-hosting
- [Full product reference for LLMs](https://www.runthru.co/llms-full.txt): everything above in one Markdown file

## Product

- [Home](https://www.runthru.co/): what it does, a real pull request comment with its video and verdict, a real bug it caught, how it works, pricing, FAQ
- [Pricing](https://www.runthru.co/#pricing)
- [Share pages, email and Slack](https://www.runthru.co/docs#share)
- [Security](https://www.runthru.co/docs#security)
- [How it works, stage by stage](https://www.runthru.co/docs#pipeline)
- [Compared with other ways](https://www.runthru.co/docs#compare)
- [Install the GitHub app](https://github.com/apps/runthru-video/installations/new)

## Examples

- [Demo repository pull requests](https://github.com/jlin3/runthru-demo/pulls): every open PR has a RunThru video, verdict and transcript
- [Pull request #2 with its walkthrough](https://github.com/jlin3/runthru-demo/pull/2)