Continue
“…but this should really be bigger.”
Component
Primary action, checkout step 2
Change
Increase size and visual weight
Someone says “make this bigger” on a call. The recording keeps the words. It loses whatever they were pointing at. Undervid keeps the sentence, the timestamp, and the frame that was on screen when it was said.
What a transcript keeps
“The spacing here is fine, but this should really be bigger. It’s the whole point of the screen.”
Which element? A week later nobody knows. Whoever builds it guesses, or asks again.
What Undervid keeps
“The spacing here is fine, but this should really be bigger. It’s the whole point of the screen.”
The primary action, at 00:12:47. Exported with the quote, the surrounding transcript, and the frame’s hash.
Illustration of an exported evidence record. Frames shown are placeholder UI.
Every moment someone pointed at something becomes a record you can act on.
“…but this should really be bigger.”
Four steps, all on your Mac
It marks what changed on screen. It won’t claim that was a cursor or a click, because nothing it ships can see one.
Claude Code and Codex connect to Undervid directly. No uploads, no copy and paste.
$ claude "implement the feedback from tuesday's review call"
→ search_meetings("checkout redesign")
→ get_context_pack(session, scope: implementation-brief)
→ read_evidence_image(00:12:47.310 · frame 22843)
3 grounded changes · 2 screenshots cited · 0 uploads
| Screen recording, audio, transcript | Never leaves |
| Coding-agent handoff | Never leaves |
| Selected frames, for AI analysis | Only when you approve it |
| Full transcript, for a summary | A separate tick box |
No analytics, no telemetry, no crash reporting. Undervid also asks you to confirm everyone on the call agreed to be recorded, and hands you a line to paste into the chat.
Free, and everything above works today. The AI that writes up the changes isn’t switched on yet, so Undervid will tell you when you reach for it.
Apple silicon · macOS 14 or later · 23 MB