Show your agent exactly
what you mean

Comms records your screen while you draw and talk. Your agent gets a brief, not a video: the region you circled, the words you said, the moment you said them.

Free during beta. $39 one time after October 1.
macOS 26 or later · Apple Silicon or Intel.

How it works

Explain it like you would to your developer

  1. 01

    Record

    Start a session. Comms captures your screen, your voice, and system audio. Whatever is on it: Figma, a staging URL, a native app.

  2. 02

    Draw and talk

    Circle the button. Underline the copy. Say what should change while you point at it. That is the whole interface.

  3. 03

    Hand it off

    Point Claude Code or Cursor at the session. It reads session.json, not a video. Then it goes to work.

Or from any terminal — the comms CLI ships inside the app: session, transcript, frames, search.

$ comms session "~/Movies/Comms/2026-08-16 21.44.47"
2026-08-16 21.44.47 — 12s, 3024×1964, 2 annotations
  a01    3.7–  5.6s  bbox 1252,673,317,295  annotations/a01.png
       “make this the primary action”
  a02    8.6– 10.4s  bbox 592,1392,256,56  annotations/a02.png
       “and move this”

$ comms search "primary action"
~/Movies/Comms/2026-08-16 21.44.47
     3.7s  make this the primary action

The brief

Your agent shouldn’t guess where you point

A screen recording is pixels. Comms sends a brief instead: one file an agent can walk through, step by step.

  • Stroke geometry with a bounding box for every annotation
  • Transcript slices bound to each stroke’s time window
  • Cropped frames of the exact region you marked
  • The app and window you were in when you drew

It is also cheaper. Forty annotations as cropped regions cost a fraction of the tokens of forty 5K screenshots, so long review sessions stay usable.

Local only

Nothing leaves your Mac

Transcription runs on-device. Sessions are folders on your disk. No account, no upload, no cloud. Use it on client work you could never paste into someone else’s server.

  • on-device transcription
  • no login, no account
  • sessions live in ~/Movies/Comms
  • works offline
  • delete a folder, it never existed

Pricing

Free while it’s in beta. Honest about what happens after

Freeuntil Oct 1 · then $39 one time
  • Every feature, no watermark, no limits during beta
  • File a real bug, or a feature request we ship, and your license is free forever
  • One-time purchase after launch. No subscription.
macOS 26 or later · Apple Silicon or Intel

Questions

Short answers

Which agents does it work with?

Anything that can read files. Claude Code and Cursor work today by pointing them at the session folder: it holds session.json, the transcript, cropped frames, and a meta.json with a howToRead note plus a bundled comms CLI (transcript / frames / search) for any shell agent. An MCP server is coming for agents without filesystem access.

Is my recording uploaded anywhere?

No. Recording, transcription, and the brief all happen on your machine. There is no server to upload to.

What does Comms need permission for?

Two: Screen Recording and Microphone. macOS will ask on first run. Both are standard for this category of app; the drawing overlay needs nothing extra.

Does transcription work in my language?

Transcription uses Apple's on-device speech models, so language support follows macOS (around thirty locales; no Norwegian yet). English works great. The first time you transcribe in a language, macOS downloads that speech model once; after that it works fully offline. More languages via Whisper are on the roadmap.

What happens when the beta ends?

On October 1 Comms becomes a $39 one-time purchase. Everyone who filed a genuine bug report during the beta — or made a feature request that made it into production — keeps a free license.