Skip to main content

The building blocks

A LiveAvatar session is a room with participants. Every session has at least two participants:
  • The end user — your customer’s browser, mobile app, or client
  • The avatar — rendered and streamed by LiveAvatar
In FULL Mode, LiveAvatar adds a third participant — our agent — which owns the full conversational pipeline (STT, LLM, TTS) and drives the avatar inside the room. You don’t build or manage this agent; it’s the core of FULL Mode. In LITE Mode, the third participant is optional and yours. You decide whether to add an agent, what it does, and where it runs. LITE Mode gives you that flexibility, but it requires you to pick an integration path up front.

LITE Mode integration paths

Five common ways to build with LITE Mode, ordered by how much you own:

1. No agent — broadcast only

Skip the agent entirely. Use LiveAvatar as a rendering layer: feed audio in, get avatar video out. Best when the end user doesn’t need real-time interaction — livestream overlays, news-style streams, narrated video generation. Your end user still joins our room to watch, but there is no conversational loop.

2. Custom agent — you build it into our room

LiveAvatar provisions the LiveKit room and the avatar; you build an agent and send it into that room. Your agent orchestrates the conversation (STT, LLM, TTS) and streams audio to the avatar. LiveAvatar exposes room access so your agent can publish audio, read events, and control avatar state. Best when you are starting from scratch and want the simplest path to a working conversation. Full control over agent logic without standing up your own transport layer.

3. Connector — we run a hosted agent for you

You provide credentials for a supported third-party voice agent; LiveAvatar stands up a connector that bridges that agent into the LiveAvatar-owned room. No agent code to write, though small configuration changes on the voice agent side are typically required to make it fully work — see the specific connector’s page for requirements. Best when you already have a voice agent with a supported provider (today: ElevenLabs).

4. Plugin — we stream into your existing transport

You already run your own LiveKit, Pipecat, or Agora stack. LiveAvatar streams avatar video into your transport instead of creating its own room. While you can build your own integration, we recommend using one of our plugins — we’ve built them with our partners to cover the most common setups. Best when you already have a real-time platform running in production and LiveAvatar is being added to an existing pipeline. Do not pick this path just to have it — path #2 is simpler if you’re not already on one of these stacks.

5. Raw transport config — you wire it up yourself

You provide transport credentials directly (livekit_config or agora_config) in the session start payload. When we send our avatar into your transport layer, you’re responsible for wiring up the integration via the WebSocket events exposed by our LITE Mode API — no plugin involved. Under the hood, our plugins are wired up this way. If no plugin exists for your stack (today: Node.js) or you’re on a transport we haven’t shipped a plugin for, this may be the path for you.
Have a specific use case or stack you’d like first-class support for? Reach out to support@liveavatar.com — we may be able to help build a plugin or connector for you.
You take on:
  • Token minting and LiveKit grants (including canPublishData: true)
  • Connecting to the LITE WebSocket returned from /sessions/start
  • Sending audio in the right format
  • Handling state and error events
Best when you can’t use a plugin (Node.js, niche transports) but still want to run your own real-time infrastructure.

Choosing a path

You have…Recommended path
An existing LiveKit, Pipecat, or Agora stack with a matching pluginPlugin — add LiveAvatar on top
An existing stack without a plugin (e.g., Node.js)Raw transport config — wire it up manually
An ElevenLabs voice agentConnector — ElevenLabs Agent Connector
Nothing yet, and you want a conversational avatarYour agent in our room
A broadcast / one-way use caseNo agent — drive audio directly

What’s next

  • Plugins — integrate with your existing LiveKit, Pipecat, or Agora stack
  • Connectors — bridge a hosted voice agent
  • Lifecycle — the three phases of a LITE Mode session
  • Events — WebSocket command and response events