Skip to main content
The Meeting Agent lets someone send their Digital Twin into a live video call. It joins as a visible participant with a face and a voice, follows the conversation, and answers when someone addresses it.
The avatar in the meeting is that person’s own twin. Their memory, their files, and their vault travel with it. It is not a shared service account, and two people sending their twins into the same call each get their own.

​
Before you begin

The Meeting Agent needs two things:
  1. The Digital Twin talks through an Anam avatar in Convo Mode — the meeting agent uses the avatar’s face and voice. See Realtime Voice & Avatars.
  2. Enable Anam Meeting Bot is on for the Digital Twin.
The meeting settings are set by Praxis AI, not by a Digital Twin’s own administrators. Sending a twin into a live call spends on outside services — the avatar, the meeting bot and live transcription — so the whole Anam Meeting Bot section is reserved to the Praxis AI team and does not appear in the twin editor for anyone else. This page describes what to ask for, and what to expect once it is on. Contact humans@praxis-ai.com to have it set up for a Digital Twin.

​
Sending the twin to a meeting

Once enabled, users ask their twin in conversation:
The twin sends its avatar into that call and reports back. Asking it to leave ends the session. Both abilities switch on and off with Enable Anam Meeting Bot — you do not manage them separately. The twin checks the link before it tries to join. It refuses Slack Huddles and GoTo Meeting links and says why: it cannot take part there as a speaking avatar. You can include a purpose and file references in the same typed or spoken invitation:
The twin keeps the invitation as background for the whole meeting, including later spoken answers and meeting chat. It is not an instruction to speak as soon as it joins. Keep it short: up to 6,000 characters are kept.
Named files must already be reachable through the twin’s normal file tools. Mentioning a file on your computer does not upload it, and an invitation does not grant access to files you cannot otherwise use. Upload or share the files first. The twin tells you when a referenced file cannot be found.
Consent is your responsibility, not the platform’s. An avatar joining a call is a visible participant, and many jurisdictions and institutional policies treat an AI participant that listens to a meeting the same way they treat recording it. Tell participants, and follow your own policy on notice and consent, before rolling this out.

​
How the twin behaves in the call

Every meeting twin behaves the same way: it listens all the time, and it speaks when someone addresses it by name (or by one of its aliases, below) or pokes it. It does not interrupt a person who is speaking. A message it posts in meeting chat follows a direct request.

​
Identity and voice

How the avatar appears and sounds in the call. Praxis AI sets these for you.
The saved Anam persona the avatar wears — its face and voice. Leave it empty to use the Digital Twin’s own avatar and voice.
Only the persona’s appearance and voice are used. Its behaviour is always your Digital Twin’s.
The participant name shown in the call — for example RAI or Touchdown Timmy. Empty defaults to Pria.
Extra names the avatar answers to when spoken aloud, comma-separated. Speech recognition often mishears a stylized name — RAI is commonly heard as Ray — so listing the variants makes it respond reliably.
Setting a display name or any alias drops the universal Pria wake word. From then on the avatar answers only to its configured name and the aliases you list. If you rename it, add every variant people will actually say.
Comma-separated written=spoken pairs that fix how a stylized name is pronounced — for example RAI=Ray, R-A-I=Ray. List each written variant explicitly.This affects spoken output only. Text the avatar posts into meeting chat is never altered.

​
Behaviour settings

A target ceiling on how long one spoken answer runs. 0 uses the platform default of 45 seconds; other values apply as 10–180 seconds. Typed chat replies are never timed.
Meetings punish long answers far more than chat does. A 20–30 second ceiling keeps the avatar conversational rather than lecturing.
An assistant the avatar loads on top of the Digital Twin’s usual behaviour — useful when meetings need a consistent specialism.The meeting avatar does not inherit whichever assistant the person happens to have active in their own chat window, so if you want a particular one in meetings, ask for it here. Empty is a perfectly good default: the twin’s own behaviour answers. Assistant IDs come from the Assistants list.The assistant follows the normal assistant permissions; it does not need to be in the one-to-one Convo picker. If it is missing, disabled or unavailable to the person who invited the twin, the twin carries on without it instead of losing its voice connection.
Which region the meeting bot runs from. The region is fixed when the avatar joins, so a change never moves an avatar that is already in a call — it applies to the next one.

​
Asking the twin to speak: Poke and Hold

Sometimes the twin should say something and nobody has addressed it by name. Type these in the meeting chat (or, on Zoom, send them to the bot as a direct message). A few things worth knowing:
  • Poke waits its turn. If the twin is already speaking, or a participant is mid-sentence, the request waits for the normal turn boundary rather than talking over anyone. It is a request for one answer, not an interruption.
  • The word has to be the whole message. poke on its own counts; “let me poke at this” is ordinary chat and the twin ignores it.
  • Names are taken at face value. Meeting platforms often show two people under one profile, so the twin uses exactly the name you typed rather than guessing from the participant list.
poke <name> is the fix for a call where several people share one screen or one account: tell the twin who it is talking to and its answers stop being addressed to the wrong person.

​
After the meeting

When the call ends, the twin posts a short recap in the meeting chat, unless Praxis AI has turned the recap off for that Digital Twin. Each recap is a summary and uses credits. What it learned can be saved to the person’s memory and vault, so it is available in later conversations. Because the avatar was that person’s own twin, the recap and anything remembered belong to them — not to a shared pool, and not to other attendees who sent their own twins to the same call. See Twin Memory.

​
Costs

Meeting sessions use credits differently from a chat turn. For every minute the meeting bot is in the call, Pria meters the avatar’s time, the meeting bot’s own time and live transcription — on top of the normal per-turn rule for whatever the twin says when it speaks.
  • The per-minute charges are what the avatar, meeting-bot and transcription providers charge, times one and a half, rounded up to a whole credit once at the end of the meeting.
  • Billing starts when the bot joins the call, not when the meeting was scheduled. Every way of ending — including asking Pria to leave — is billed the same.
  • If the Digital Twin pools credits, the pool pays; otherwise the individual’s balance does.
What drives the bill is mostly time in the call: for a quiet hour where the twin barely speaks, the per-minute charges are the bulk of it. See Credits for the full rules.
If your organization supplies its own Anam credentials, Anam bills that account directly for the avatar’s time, so that charge does not appear in your Pria credits. The meeting bot and transcription are always run by Praxis AI, so their per-minute charges still bill in credits.
Before rolling this out widely, set per-user limits under User Quotas (Instance Settings › Quotas) so one enthusiastic pilot user cannot drain a shared pool. See Credit Management.