SDKs

Embed a real-time, lip-synced bitHuman avatar natively in your app — on-device or over the cloud — from Python, Apple, Android, or the browser.

One engine, one API, every language

Every bitHuman SDK is a thin, idiomatic binding over libessence — the portable C++ avatar-rendering engine that also powers the CLI and the cloud REST API. They all read the same .imx avatar file and produce identical frames at 25 FPS, so anything you prove out in one language ports straight to the others.

The contract is the same everywhere:

Push 16 kHz mono PCM audio as it arrives (mic, TTS, WebRTC).
Drain lip-synced video frames at 25 FPS.

That push/drain loop is documented once, canonically, under audio streaming — every language page links back to it rather than repeating it.

Which SDK should I use?

Match your target platform to a binding. All four ship visual, lip-synced avatars; the difference is where they run and how mature each surface is.

Platform / target	SDK	On-device / cloud	Status
Backend, AI agents, batch render, edge boxes	Python	On-device	GA
Native Mac, iPad, iPhone apps	Swift / Apple	On-device	GA
Native Android apps	Android (Kotlin)	On-device	Beta
Browser & Node web clients	JavaScript / TS	Cloud client	Preview
Native app ↔ server avatar over WebRTC, or Python voice agent with a face	LiveKit (Apple + Python)	Cloud / hybrid	Built on the GA bindings

If you are not sure, start with the Python SDK or the no-code CLI.

On-device vs cloud

bitHuman runs in two topologies. The same .imx and the same API work in both.

On-device
Runs on the user's machine, phone, or edge box. Private, low-latency, offline-capable. Python, Swift, Kotlin.

Cloud
We host the GPU. Zero ops, ideal for web clients and sharing one avatar. JavaScript/TS, LiveKit.

	On-device	Cloud
Where inference runs	Your machine / phone	bitHuman’s GPU pool
SDKs	Python, Swift, Kotlin	JavaScript/TS, LiveKit
Network	Optional — billing heartbeat only (or fully offline)	Required
Hardware	Any modern CPU (Essence); Apple Silicon / NVIDIA GPU (Expression)	None — we host it
Cost	1 credit/min (Essence) · 2 credits/min (Expression)	2 credits/min · 4 credits/min
Best for	Privacy, kiosks, edge, low latency	Zero-ops, web clients, sharing one avatar

See models for the Essence vs Expression comparison and pricing for credit details.

Status matrix

We keep this honest so you can plan around it.

SDK	Package	Topology	Status
Python	`pip install bithuman`	On-device	GA
Swift / Apple	SwiftPM `bitHumanKit`	On-device	GA — `bitHumanKit` is the published full-stack package; the newer `libessence` `Bithuman` streaming binding is rolling out
Android (Kotlin)	Gradle `ai.bithuman:sdk`	On-device	Beta
JavaScript / TS	`npm install @bithuman/sdk`	Cloud client	Preview
Flutter	reference app only	On-device	Reference app in `bithuman-apps`, not a published code SDK — see below

Note Swift has two surfaces today. The currently-published SwiftPM package is bitHumanKit — a full-stack on-device library with a built-in voice agent (STT + LLM + TTS) and a low-level streaming runtime. A newer binding named Bithuman maps directly onto the libessence streaming engine (Fixture / Runtime / pushAudio / pullFrame) and is rolling out. The Swift page covers both.

A note on Flutter

A Flutter integration exists, but it is a reference app in the private bithuman-apps repo — it is not published to pub.dev and the plugin source is not in a public repo. Do not add bithuman: ^X.Y.Z to a pubspec.yaml; flutter pub get will fail with “package not found”. Until it ships, build Flutter apps on the underlying Swift SDK (Apple) and Android SDK via platform channels, or ping us on Discord for early access.