Product

Everything voice, from one Control API

The OneSay Platform separates what your client may do from what it asks to do. These are the services and clients it is built around.

Task voice

Your application submits caller-supplied audio to an explicit Task. The Realtime WebSocket carries it with sequence numbers, acknowledgements, backpressure, replay and resume, and the Task produces its result and Artifacts.

Caller-supplied audio in, Task result out
Acknowledged, resumable delivery
Results and usage you can read back

Realtime RTC

Rooms carry live audio and data between participants. OneSay decides who may join, publish and subscribe, and changes to those rights reach live sessions.

Audio publications and subscriptions
Data packets, streams and RPC
Permission changes applied to live sessions

Agents

An Agent works on the publications you select, nothing more. You create, read and stop each execution explicitly.

Explicit input publications
Create, read and stop by identifier
Results kept separate from Task dictation

SDKs

Two SDKs on the same server model, with each SDK's capabilities in the compatibility matrix. Both are private candidates until the public release.

JavaScript SDK for browsers and Node.js
ESP32 SDK as an ESP-IDF component
A documented capability and compatibility matrix

Security model

OneSay holds the authority even when a client is modified or bypassed: scope comes from the server, credentials are short-lived, and sensitive actions need recorded Consent.

Pre-release: the OneSay Platform and its SDKs are not publicly released. Nothing on this site is an offer of service.