Skip to main content

Record a task on a system

Record yourself doing a task on a system: the clicks, what you typed, screenshots, and your narration. You then review the recording and hand it to an AI agent. Because the "recording" is a collection of screenshots and descriptions in text, AI agnets are

Record yourself doing a task on a system: the clicks, what you typed, screenshots, and your narration. You then review the recording and hand it to an AI agent. Because the "recording" is a collection of screenshots and descriptions in text, AI agnets are considerably better at ingesting compared to a direct video.

Before you start

  • A system that's online. Offline systems can't be recorded.

  • A clear idea of the one task you'll record. Keep it short, because recordings stop after 30 minutes.

Step 1 - Start the recording

  1. Go to Recording and click New recording.

The New recording dialog with System, hidden offline systems, Record narration and Start recording numbered 1 to 4
  1. Choose the System.

  2. Offline systems aren't listed. The dialog says how many are hidden.

  3. Turn on Record narration to talk through what you're doing. Your speech is transcribed in the browser. No audio is uploaded or stored.

  4. Click Start recording. The viewer opens in a new tab.

Step 2 - Do the task

Do the task in the viewer tab. The recording's page shows its progress:

A recording in progress with its status, Refresh, counters, Stop recording and Narration numbered 1 to 5
  1. The status shows Recording.

  2. Refresh updates the counters.

  3. Elapsed, Events and Captured. Only elapsed time updates on its own, so zero events part-way through is normal.

  4. Stop recording, and Open viewer if you've closed the viewer tab.

  5. Narration. Click the microphone to talk, or type a line in Note. Each line is added at that point in the recording.

Step 3 - Stop the recording

When you've finished the task, click Stop recording. The recording moves to review. See Review and approve a recording.

Tips for narrating

The narration is what explains why you did each step, and an AI agent relies on it most.

  • Say what you're about to do, then do it. "I'm opening Installation to create a key."

  • Say why. "I'm choosing the staging key because it files the system in the right folder."

  • Name things as the reader would. Say "the Systems screen", not "over here".

  • Mention anything that could go wrong. Warnings and prerequisites are what a written procedure needs most.

  • Say when you've made a mistake. "That's not what I meant to click", then exclude it during review.

NOTE: A recording captures everything on screen and everything you type, including passwords and keys. Review it before you approve it. See Review and approve a recording.

Who can record

Recording is controlled by the session.record capability, one of the six risky capabilities. If an API key's capability policy gates it, a person has to approve before that key can record. See Create and use capability policies.

Check it worked

  • After you stop, the recording shows Ready for review.

  • Admin > Activity Log shows recording.started and recording.stopped.

Related articles

Did this answer your question?