Skip to main content

​
Overview

When you invite an agent to a Zoom, Microsoft Teams or Google Meet call, it can share its screen and show you what it finds. Its screen is a live canvas. The agent places things on it and changes them in place while it talks:
  • Charts, drawn live from query results, such as a bar chart of games per year.
  • Tables that can be sorted and paged.
  • Diagrams, written in Mermaid.
  • Cards with formatted text and tables.
  • Headlines, notes and simple layouts.
  • Pictures from the meeting’s own files.
The agent also draws on the canvas: a pointer, a box, a circle, an arrow with a label, a freehand line or a short note. It can highlight one bar, zoom into a range, or sort a table. These changes update the same chart; it is never replaced by a new picture.

​
Asking before it shares

The first time the agent wants to show something in a meeting, it asks once, out loud and in the meeting chat:
I can show that to you — do you mind if I share my screen?
  • Anyone in the call can answer, by voice or in the chat. A plain “yes”, “sure”, “go ahead” or “I don’t mind” lets it share. The meeting’s transcript records who answered and how.
  • “No”, or no answer within a minute, means it doesn’t share. It offers to describe the result instead, and it asks again only after someone asks to see something.
  • After a yes, it doesn’t ask again or count down before each update. New charts, highlights and marks just appear.

​
One screen share for the whole meeting

The agent starts its screen share once and keeps it for the rest of the meeting. Each new chart, edit or mark changes the shared screen in place, so the screen never goes blank and comes back between views. The share ends when:
  • someone says “stop sharing”, or types /<agent name> stop in the chat; or
  • the agent leaves the call.
After a stop, the agent asks again the next time it wants to share.

​
Asking for what you want to see

Talk to the agent the way you would talk to a colleague who is sharing their screen:
  • “Graph games per year for the 1900s and highlight 1904.”
  • “Circle the March bar.”
  • “Show that as a table, sorted by revenue.”
  • “Zoom in on 2019 to 2021.”
  • “Put a note next to the spike.”
  • “Undo that” or “clear the marks.”
  • “Draw the pipeline as a diagram.”
The agent’s live voice changes the canvas while it speaks. Longer work, such as running a new query or building a chart from scratch, happens in the background in the agent’s meeting thread. The result appears on the canvas as soon as it is ready, and the agent talks you through it.

​
Asking for data in the call

The person who invited the agent, or someone they confirmed with a code, can ask it for data directly:
  • “What databases are in Databricks?” lists them on screen as the agent talks.
  • “Show me the first rows of the races table.”
  • “How many races per year? Chart it.”
These quick reads run as the person who invited the agent, with their access. Each is one read-only query of at most 200 rows and 30 seconds. A placeholder appears at once and fills in as the rows arrive. Anything that changes data, or takes longer, goes to the agent’s background work. Its results appear on the canvas as they come in: rows first, then the chart. Other participants can still ask to see what’s already on the canvas, but they can’t ask the agent to fetch data.

​
What it can show

The canvas holds only what the person who invited the agent may see:
  • Query results from the meeting’s own conversation, and only when the inviter may read that conversation.
  • Files from the meeting’s own conversation, under the same rule.
The agent shares nothing in a meeting that no Genesis user invited it to.

​
If the screen share can’t start

Some meeting bots can’t share a screen. In that case, the agent shows the canvas on its own video tile and says so in the meeting chat.