Glossary
The words these docs use, each in a sentence or two, with a link to where it is explained in full.
Every term has its own link: /docs/v2/getting-started/glossary#avatar, /docs/v2/getting-started/glossary#credit and so on.
API key
A secret your server sends with every API call, in the X-Api-Key header. It belongs to your
Atmee account: what it may do is set by its scopes, what it may spend by its
spend cap. It must never reach a browser: a key in page JavaScript can be read by every visitor. Create
one in Studio → API. See API keys.
Avatar
A talking video character, built on one model. It is made from a portrait, a voice sample and a persona, and can answer from knowledge files. Each avatar belongs to one account and has an avatar ID. Make one in Studio or with Create an avatar.
Avatar ID
The identifier of one avatar, passed as avatarId when you start a
session. In Studio it's the UUID in the avatar's editor address,
/studio/avatars/<id>/edit, and the API returns it from POST /v1/avatars.
Avatar mode
The mode where you bring your own voice agent, with its own speech recognition, language model and
voice, and Atmee renders only the avatar's face, in sync with your agent's audio. Start
it with the LiveKit Agents plugin or POST /v1/avatar_sessions. It costs
half of Behavior-driven Full-stack.
Avatar session
A session in Avatar mode: the avatar joins your LiveKit room, or one Atmee hosts, and speaks your agent's audio. See the LiveKit Agents plugin.
Behavior-driven Full-stack
The mode where Atmee runs the whole conversation: behavior (powered by Causeur), the
language model, speech recognition, the voice and the face. Your side is one server endpoint that
calls POST /v1/sessions and a page that joins the room. See the
Quickstart.
Behavior-Driven Voice Mode
The mode where Atmee runs the conversation with Causeur and the avatar's voice, with no
video. Your server calls POST /v1/speech_sessions and your client streams audio over a WebSocket.
Causeur
Atmee's behavior model. It decides how an avatar acts, not just what it says: personality, emotion, timing and active listening. It runs in Behavior-driven Full-stack and Behavior-Driven Voice Mode.
Credit
The unit everything is paid in. On an Atmee V1 model, a Session API session costs 1 credit per minute and an avatar session 0.5; Atmee V2 costs twice that. Your plan includes credits each month, and you can buy more under Studio → Billing. See Costs.
Knowledge
Documents an avatar answers from: PDFs, Word files, text, Markdown, HTML or CSV, up to 20 files. Add them when you create the avatar, or replace them later.
LiveKit
The real-time video and audio service a conversation runs over. With the Session API Atmee runs it for you; with the LiveKit Agents plugin you use your own LiveKit project.
LiveKit Agents plugin
A Python package for developers who already run a voice agent on LiveKit Agents. Your agent decides what to say; the plugin adds an Atmee avatar that says it. See the LiveKit Agents plugin.
Model
The AI that renders an avatar's face: Atmee V2, our latest and most human-like, or
Atmee V1, realistic at half the credits. An avatar's model is set when it is created
(engine in the API). See Models.
Persona
Who an avatar is: its name, personality, greeting, how it introduces itself, and the rules it must not break (guardrails). The persona and the knowledge decide what the avatar says.
Plan
Your Atmee subscription: Free, Starter, Growth, Pro, Scale or Enterprise. It sets the credits you get each month, the longest session and how many sessions run at once. API access starts on Starter, and Atmee V2 on Growth. See the pricing page.
Portrait
The image an avatar's face is rendered from: one photo or picture showing the whole head with some margin around it.
Room
The live video call a session takes place in. The avatar and your user each join it as a participant. Your page joins with the server URL and user token the session returns.
Scope
A permission on an API key. sessions lets it start sessions, avatars:read and
avatars:write let it read and manage avatars. Give each key only the scopes it needs. See
Scopes.
Session
One live conversation between one user and one avatar. Your server starts it with
POST /v1/sessions; it ends when the user leaves or its time runs out, and it is billed by the
minute in credits. One avatar holds one session at a time.
Session API
POST /v1/sessions, the endpoint that starts a
Behavior-driven Full-stack session.
Spend cap
A limit on how many credits one API key may spend, in total, per month or per day. When it is reached, new sessions with that key are refused. See Caps.
Studio
Atmee's web app at atmee.ai/studio, where you create and manage avatars, API keys and billing. Everything you do with the API, you can also see there.
User token
What your page needs to join a session's room. POST /v1/sessions returns a
serverUrl and a userToken, a short-lived token good for that one room only. Your server
passes both to the page and keeps the API key to itself.
Voice sample
A recording of the voice an avatar should speak with: 10 to 60 seconds of clear speech from one person.
Building with an AI agent? The agent skills, the MCP server and the atmee CLI are explained
on their own pages.
Docs for AI agents
Every page of these docs, and the API's contract, as plain files an AI agent can read: an index, the whole docs in one file, a Markdown copy of each page, and the OpenAPI spec.
Avatar 101: Negative examples
The portraits that animate badly, and what to use instead. Style and framing are never the problem: Atmee V2 takes half-body, full-body and even squatting shots.