01 / DECIDE
Architecture decision
Use it for low-latency speech or multimodal interaction where request-response audio would feel discontinuous.
Transport, session configuration, turn detection, interruption, tools, reconnect, and terminal states jointly define the experience. Map input, output, state, and side effects as one observable path before deciding which layer owns design realtime as a live distributed system.
02 / BUILD
Three-step implementation
- 01
Frame the contract
Write down the caller, data classification, success condition, timeout, cancellation, and ownership. Use it for low-latency speech or multimodal interaction where request-response audio would feel discontinuous.
- 02
Build one narrow path
Implement one end-to-end path with request correlation, typed state, and reversible failure handling. Transport, session configuration, turn detection, interruption, tools, reconnect, and terminal states jointly define the experience.
- 03
Prove the outcome
Turn acceptance into a repeatable fixture, contract test, or browser test. Network tests cover jitter, reconnect, double-talk, device changes, long sessions, late tools, and clean shutdown.
03 / BOUND
Production boundary
Client credentials must be short-lived, interruptions must cancel stale work, and tool results need turn ownership.
04 / PROVE
Acceptance evidence
Network tests cover jitter, reconnect, double-talk, device changes, long sessions, late tools, and clean shutdown.
SOURCE / HTTP
Reproducible source probe
curl -fsSI 'https://developers.openai.com/api/docs/guides/realtime' | sed -n '1,5p'