> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vibeconf.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Working together in a call

> Two things you can tell a bot out loud that make it noticeably better in a room.

Two habits that make a bot in your meeting better, neither of which requires touching a
setting. Both come out of running this in real rooms, and both are **things we are trying**,
not settled practice.

## Correct it out loud, in the moment

The most useful feedback you can give a bot is the feedback you'd give a person, said at the
time — not written down afterwards.

The problem is that "that wasn't great" is ambiguous. Did the bot say the wrong *thing*, or
did it say a fine thing at the wrong *moment*? Those are different problems.

So we're trying a marker word for the second kind:

> **"Awkward — Pepper, you cut in while I was still talking."**
>
> **"Awkward — Jimmy hasn't said anything in five minutes."**

**The shape is: `awkward` · who · why.** Naming who matters as soon as there's more than one
bot in the room. The *why* is what makes it a correction rather than a complaint.

Say it out loud, in the call. Your bot hears the room through live captions, so it arrives as
ordinary speech, in sequence, next to the thing it's about.

<Note>
  **Pick a word you don't otherwise use**, so it can't be confused with ordinary conversation.
  We picked "awkward." Captions do mis-transcribe words, so check that yours survives before
  relying on it.
</Note>

**Mark the good moments the same way.** If the only signal a bot gets is *awkward*, the safest
way to never be awkward is to never speak. We have seen the same bot described as
over-interrupting in one session and under-interrupting in the next, which is how narrow the
target is. So: *"nice — that was the right time to jump in"* beats "thanks."

<Warning>
  This is a **convention we are experimenting with, not a product feature.** Nothing in the
  app watches for the word. It works only because the bot hears it like any other speech and
  its own instructions tell it to take the note. If you want it, put it in your bot's
  instructions.
</Warning>

## Tell it what kind of room it's in

A bot reads the room through Google Meet's live captions, and caption attribution is **per
participant, not per voice**.

So if eight people are sitting around one laptop, the caption stream shows the bot one person
talking. It can't tell them apart from what it hears.

Which means the single most useful thing you can say in the first minute is:

> **"There are eighteen of us in a conference room on one mic, plus six people remote."**

A bot that knows it's listening to a shared room mic can hold a higher bar before speaking,
because in that setup a cough, a chair, or a side conversation all arrive looking like someone
taking the floor. What it does with that is up to your bot — but it can't act on what nobody
told it.

In our own sessions, the rooms that have gone worst are the ones where many people shared a
single microphone. It's worth saying out loud when that's your setup.

## About the whiteboard

Boards may still be there when you come back to the same room. **Retention isn't guaranteed
and may change** — so if a board matters, save or export it with your call notes before you
leave.

<Note>
  For what the server actually guarantees about board storage and recovery, see the
  Vibeconferencing app's own documentation — that's the authoritative source, and it's kept
  next to the code so it stays true.
</Note>