Runnable recipes for Synthesia Interactive Avatars: photoreal, lip-synced avatars you can talk to in your browser.
Each recipe is self-contained, with its own README, dependencies, and .env.example, so you can copy a directory out and build on it directly.
| Recipe | What it shows |
|---|---|
| minimal | The smallest useful app: a Python LiveKit voice agent (STT → LLM → TTS) with the avatar integration in three lines. Start here. |
| rag | A LiveKit agent grounded in a knowledge base (public Wikipedia out of the box, AWS Bedrock for your own corpus). It speaks only from retrieved source material. |
| tools | LLM uses tools to fill a booking form in the browser as you speak, and your edits flow back to the agent: tool calling plus UI state sync over LiveKit data channels. |
Prerequisites
You need two things before you start:
-
A Synthesia API key with Interactive Avatar access. This authenticates you to the Interactive Avatar API.
-
The Synthesia plugin, installed via pip:
pip install "livekit-plugins-synthesia~=1.8"Each recipe's README lists everything else it needs on top of these: providers, keys, and how to run it.