Glossary

The words used across the bitHuman docs, what each one means, and the older names you may still meet in code, files and saved links.

What each term means here, and the page that owns it.

Models and avatars

TermMeaning
Essence 2 (essence-2)The model that renders a photoreal person from one portrait. Essence 2
Expression 2 (expression-2)The model that renders any character from one portrait. Expression 2
Essence 2 MaxEssence 2 Max is available on the Enterprise plan only. Contact sales to enable it. Models
Essence 1, Expression 1 (essence-1, expression-1)The first generation of the models. First generation
AgentAn avatar together with its persona, voice and knowledge, created in the bitHuman app or with the Agents API.
Agent codeThe identifier of an agent, for example A23WJF0199. Agent codes
Sample avatarsPublic agents anyone can use without an account: sofia-ramirez (Essence 2) and wise-pup (Expression 2). Showcase list
Avatar file (.imx)One agent’s model files in one container, downloaded once and rendered on your hardware. The avatar file
PersonaThe agent’s system prompt: who it is and how it answers. Persona
GesturesA named motion (wave, nod, clap) an avatar plays when your code asks; the API calls them dynamics. Gestures
IdleThe motion an avatar shows between replies, so it never freezes.

Where it runs

TermMeaning
Renders / runsThe avatar renders (on the device, in the browser, on your server or in the bitHuman cloud); the conversation runs (in your stack, in the CLI’s local conversation brain, or on bitHuman’s servers).
On the deviceThe avatar renders on the user’s iPhone, iPad, Mac or Android phone. On the device
CPU only (no GPU)The avatar renders on a standard Linux PC with no graphics card. CPU only
Your serversYou run the CLI, the Python SDK or the LiveKit plugin on machines you control. Your servers
bitHuman cloudThe avatar renders on bitHuman’s servers and streams to any screen. bitHuman cloud
Fully offlineA license for realtime avatars with no internet connection. Fully offline
File renderingTurning an audio file into a talking-avatar video with bithuman render or the Python SDK. It signs in online; it is not the offline license. Talking video
WebGPUThe browser graphics interface the web embed uses to render the avatar in the tab (render=local). Web
Local conversation brainThe CLI mode (BITHUMAN_LOCAL=1) that runs speech recognition, the language model and the voice on a Mac or Linux PC. Local conversation brain
× real timeSeconds of avatar video rendered per second. At 1.0× or more an avatar holds a live conversation. Performance

Credentials and billing

TermMeaning
API secretThe one credential for every surface, read from BITHUMAN_API_SECRET; REST sends it in the api-secret header. Your API secret
Embed tokenA short-lived token your server mints so a web page can open a private agent without the secret. Embedding
Runtime tokenA short-lived token for a LiveKit room or a download, minted with the API secret. Authentication
CreditsWhat usage is paid in. Pricing and credits
Active session timeWhat realtime usage bills: the time a session is open, talking or idle, to the second. Pricing
Creator planThe plan API and SDK use requires, or a higher one. Plans

Older names

The model names retired over time, the older file extensions and library names, and what each means today, are in one table: Naming & migration. The SDK and API names below are older spellings you may still meet:

Older nameWhat to use now
BITHUMAN_API_KEYBITHUMAN_API_SECRET. The deprecated alias is still read, with a warning, until CLI 3.0 and bithuman 4.0.
POST /v1/realtime/ephemeral-token (ek_… tokens)Retired; connect through the realtime relay.
bithuman.offline, render_offlineDeprecated; use bithuman.open(path).render(audio, out_mp4=...) (Python).
bitHumanKitA legacy Swift package, not the current one; use the Swift package products Expression2 and Essence2Kit.

Next