Integration boundary
What connects to what
The public Backend Mode server uses FastAPI WebSockets around an application-owned ASR, LLM, TTS stack and the Python Spatius Server SDK.
Prerequisites
- Python 3.10+
- Spatius Python Server SDK credentials
- A defined downstream transport such as LiveKit or an owned WebSocket
- A production secret manager and session authorization model
Implementation path
Build a testable baseline
- 1
Start from the official Backend Mode Python server reference.
- 2
Create a FastAPI lifespan that initializes shared SDK resources once per process.
- 3
Expose an authenticated session endpoint and keep provider secrets server-side.
- 4
Stream assistant TTS audio into the Spatius session and forward synchronized audio and motion through the chosen transport.
- 5
Handle disconnect, cancellation, backpressure, and session cleanup before load testing.
Decision guidance
Best for and not best for
Best for
- Python voice-agent backends
- Enterprise teams that require server-side policy and audit controls
Not best for
- Frontend-only prototypes
- Teams that do not want to operate session infrastructure
Evidence-aware assessment
Advantages and limitations
Advantages
- Fits existing Python AI orchestration
- Keeps secrets and policy enforcement on the server
- Allows explicit transport and observability choices
Limitations
- You own scaling, reconnect behavior, and downstream transport
- A web framework does not solve audio pacing or motion synchronization by itself
- No first-party FastAPI concurrency benchmark is present in the supplied product pack
Do not confuse this runtime with a small Direct Mode token endpoint. You own per-session backpressure, provider secrets, concurrent disconnects, and cleanup. Indexing this guide does not convert an architecture blueprint into a claimed native connector; run and document the validation checklist before production use.
Page-specific FAQ
Implementation questions
Is FastAPI required by Spatius?
No. FastAPI is one application-server option around the Python Server SDK.
Can the API return finished avatar video?
The documented Spatius architecture produces synchronized audio and motion for client-side rendering; it is not described as a finished-video streaming endpoint.
What must be load tested?
At minimum: concurrent sessions, audio backpressure, disconnect cleanup, transport reconnects, memory, and end-to-end failure recovery.
Primary evidence
Sources and verification
Reviewed 2026-08-10. Recheck versions and implementation details before publication.
Continue exploring