THE OLD WAY
Re-watch the whole clip
14 minutes of footage to find the 40 seconds that mattered.
SCREEN RECORDING → SPOKEN FINDINGS
screenscribe
Talk through the bug once. Get a fix-ready report.
Record your screen and say what's wrong. screenscribe pulls out the important moments, captures the matching frames, and writes the report for you.Use it yourself, send it to a teammate, or hand the Markdown/JSON to a coding agent.
$
✓ transcript: 42 segments
✓ findings: 3 bugs, 1 change
✓ report.html report.md report.json
THE PROBLEM
A recording isn't a bug report
Hitting record and talking through what's broken is the fastest way to show it. But a video isn't actionable - someone still has to watch the whole thing, scrub to the right second, and type up what was said. That someone is usually future-you, at the worst possible time.
THE OLD WAY
Scrub, pause, screenshot
Frame by frame, hunting for the exact moment you meant.
THE OLD WAY
Write it all up by hand
Turn "this bit" into words a teammate or agent can follow.
05:3414:02
"...okay so if I click here it just... nothing happens, and then the total up top is still wrong, let me show you again..."
demo_report.html
EXECUTIVE SUMMARY
3 bugs, 1 UI change. Highest impact: checkout total not recalculating after item removal.
BUG
Total not recalculating
Removing an item leaves the cart total unchanged.
UI
Button has no click state
"Add" gives no feedback - users tap twice.
ANATOMY OF A FINDING
One marked moment. Fully written up.
Every finding carries the proof: timestamp, transcript, matching frame, visual check, and a concrete suggested fix.
Finding #03
MEDIUM
"This button looks clickable, but nothing happens when I press it..."
WHAT SCREENSCRIBE NOTICED
The primary CTA appears enabled, but the user reports no response on click.
SUGGESTED FIX
Check the click handler and the disabled / loading state on the CTA component.
- AND IT SHIPS AS THREE FILES
report.html
Open in your browser - synced player, subtitles, and screenshot annotations. No server.
report.md
Paste straight into your coding agent - no video, just what to fix.
report.json
Feed into your own tooling, dashboards, or issue tracker.
HOW IT WORKS
Record. Detect. Verify. Export.
The workflow follows the path you actually need: capture the issue, let screenscribe pull out the moments, verify the frame, and export a report.
Recordscreen plus narration
→
Detecttranscript finds the moments
→
Verifyframes confirm the context
→
ExportHTML · Markdown · JSON
The transcript tells the analysis what to look for before the frame check runs.
Record
Capture your screen and narrate the bugs and changes in plain language. Any format FFmpeg reads works - mp4, mov, mkv, webm.
Detect
screenscribe transcribes the narration and pulls out the moments that sound like bugs, changes, or UI issues.
Verify
Each moment is checked against the matching frame, so the report is grounded in what was actually on screen.
Export
Open the HTML, send the Markdown, or pipe the JSON into your own tooling or coding agent.
PICK YOUR MODE
Which mode should I use?
Start with review when you narrated the walkthrough. Use analyze when the clip is silent, messy, or needs careful manual marking.
review
For narrated walkthroughs
Run one command and get a full report: transcript, detected moments, screenshots, visual checks, and exports.
transcribefind momentsscreenshotconfirmreport
analyze
For silent clips or careful QA
Open the local dashboard, mark moments yourself, add notes, and analyze only the frames that matter.
open dashboardmark framesadd notesanalyze pickedexport
FEATURES
Built for a clean handoff
Everything the report needs travels with it, so a finding is ready to act on the moment it lands - for a teammate or an agent.
One interactive file with a synced player, subtitles, and screenshot annotations. No server, no external requests - open it, email it, keep it forever.
review runs the whole pipeline for you; analyze hands you a dashboard where you mark the moments yourself.
Scrub the video, flag the frames that matter, and drop a voice or text note - so the AI only looks where you point it. Works even with no audio track.
A transcript-first lane produces clean TXT, segments JSON, and WebVTT - filtered for hallucinations, ready to paste straight into an agent.
Point screenscribe at OpenAI or any compatible provider and use your own key. You pay the provider directly - there's no screenscribe account, subscription, or charge in between. Vetcoders ships LibraxisAI as the first-party default.
WHO IT'S FOR
Show it once - skip the write-up
You don't have to be deep in the codebase. If you can record your screen and say what's wrong, screenscribe does the write-up.
DEVELOPERS
Self-review before you push
Walk your own change, catch what's off, and get a checklist you can work through.QA
Bug demos, already written up
File a repro that's a report, not a raw clip the dev has to decode.PRODUCT
Feedback walkthroughs
Record a walk-through of the build and turn your notes into structured asks.VIBECODERS & AGENTS
Hand it straight to your AI
The Markdown is built to be read by an agent - say what's wrong, drop the report in, let it fix.INSTALL
Get started in three steps
Clone and install, add your key, run a review. Bring your own OpenAI-compatible key - you pay the provider directly, nothing in between.
Runs locally. Uses your provider key.
screenscribe processes the video on your machine and only sends configured AI requests to the endpoint you choose. No screenscribe account, no subscription layer.
Clone & install
Grab the repo and run make install. It installs only the CLI and runtime dependencies via uv tool — no contributor toolchain or Git hooks.
Choose your provider
Run the setup wizard for LibraxisAI, OpenAI, or a custom compatible endpoint. Your API key is entered through a hidden prompt.
Run a review
Install FFmpeg first (brew install ffmpeg on macOS or sudo apt install ffmpeg on Debian/Ubuntu), then point screenscribe at a recording. On a shared Mac, ask the Homebrew owner or an administrator to install it; do not change ownership of the Homebrew prefix.
BASH
$ git clone https://github.com/vetcoders/screenscribe.git
$ cd screenscribe
$ make install
# choose a provider and enter your key in a hidden prompt
$ screenscribe config setup
# FFmpeg is required before the first video run
# review a narrated recording
$ screenscribe review demo.mov
Python 3.11+
FFmpeg
uv
make