How synthesia-ai/interactive-avatar-quickstarts works
This repository is a set of three runnable browser-based quickstarts for Synthesia Interactive Avatars: a minimal spoken assistant, a knowledge-grounded assistant, and a booking-form assistant. In each recipe, a local host serves the page and issues a LiveKit room token that dispatches its voice agent; the agent connects speech/model services and a Synthesia avatar. The RAG recipe retrieves source passages before answering, while the tools recipe synchronizes form edits in both directions. Connections for unsampled browser pages and some provider wiring are limited to the README or explicit agent comments.
Main entry points: rag/agent.py Grounded voice agent · minimal/agent.py Voice agent · tools/agent.py Form voice agent · index.html Booking form page
Components 23
Minimal voice demo
- Room token host
minimal/server.py - Avatar page
minimal/index.html - Voice agent
minimal/agent.py - Realtime preflight
minimal/realtime_preflight.py
Grounded answers
- Room token host
rag/server.py - Avatar page
rag/index.html - Grounded voice agent
rag/agent.py - Query rewriting
rag/agent.py - Wikipedia retrieval
rag/agent.py - Bedrock retrieval
rag/agent.py - Passage-grounded context
rag/agent.py - Pipeline status
rag/agent.py
Booking assistant
- Room token host
tools/server.py - Booking form page
tools/index.html - Form voice agent
tools/agent.py - Form operations
tools/agent.py - Form data sync
tools/agent.py
Realtime platform
- LiveKit rooms
- Speech and LLM services
- Synthesia avatar
- Wikipedia API
- AWS Bedrock KB
Other components
- Browser visitor
Connections 35
- Browser visitor → Avatar page: uses
README.md - Room token host → Avatar page: serves
minimal/server.py - Room token host → LiveKit rooms: mints room token
minimal/server.py - Room token host → Voice agent: dispatches worker
minimal/server.py - Voice agent → Realtime preflight: installs support
minimal/agent.py - Voice agent → LiveKit rooms: starts voice session
minimal/agent.py - Voice agent → Speech and LLM services: uses speech and model
minimal/agent.py - Voice agent → Synthesia avatar: starts avatar session
minimal/agent.py - Browser visitor → Avatar page: asks questions
README.md - Room token host → Avatar page: serves
rag/server.py - Room token host → LiveKit rooms: mints room token
rag/server.py - Room token host → Grounded voice agent: dispatches worker
rag/server.py - Grounded voice agent → LiveKit rooms: starts voice session
rag/agent.py - Grounded voice agent → Speech and LLM services: uses speech and model
rag/agent.py - Grounded voice agent → Synthesia avatar: starts avatar session
rag/agent.py - Grounded voice agent → Query rewriting: rewrites query
rag/agent.py - Query rewriting → Speech and LLM services: requests rewrite
rag/agent.py - Grounded voice agent → Wikipedia retrieval: retrieves passages
rag/agent.py - Wikipedia retrieval → Wikipedia API: searches and extracts
rag/agent.py - Grounded voice agent → Bedrock retrieval: retrieves passages
rag/agent.py - Bedrock retrieval → AWS Bedrock KB: queries knowledge base
rag/agent.py - Grounded voice agent → Passage-grounded context: injects retrieved text
rag/agent.py - Grounded voice agent → Pipeline status: publishes stage
rag/agent.py - Pipeline status → Avatar page: updates status display
README.md - Browser visitor → Booking form page: edits booking form
README.md - Room token host → Booking form page: serves
tools/server.py - Room token host → LiveKit rooms: mints room token
tools/server.py - Room token host → Form voice agent: dispatches worker
tools/server.py - Form voice agent → LiveKit rooms: starts voice session
tools/agent.py - Form voice agent → Speech and LLM services: uses speech and LLM
tools/agent.py - Form voice agent → Synthesia avatar: starts avatar session
tools/agent.py - Form voice agent → Form operations: registers form tools
tools/agent.py - Form operations → Form data sync: publishes field changes
tools/agent.py - Form data sync → Booking form page: syncs form state
tools/agent.py - Booking form page → Form data sync: sends edits
tools/agent.py