Change Log

[2025.12.07.1623] - 2025-12-07

Added

ClojureScript

A new event of type "reelthyme.call_id" will be sent to WebRTC channels. This event is used to capture the call id of the session which can be used for sideband channels on the server.

Clojure

The connect! method no accepts a :call-id parameter that allows server side control of a client side conversation. If :call-id is given, :model is ignored. If using a :call-id, ensure that an :api-key is given using the :value of the client-secret generated for authenticating the client side channel.

Changed

BREAKING CHANGES

• The WebRTC transport used via CLJS no longer uses js->clj by default. Native JS objects are the default. js->clj can still be applied by providing an :xf-out transducer.
• The websocket transport no longer supports a :uri param, and instead supports :model and :call-id. :model defaults to gpt-realtime but can be changed to gpt-realtime-mini for example.

[2025.11.30.1359] - 2025-11-30

Added

ClojureScript Only

A :media-stream-track param is now supported. Supersedes :content-types if provided. Useful for workflows requiring more nuanced permission flows - for example calling navigator.mediaDevices.getUserMedia() manually and then connecting only after permission is granted.

(connect! client-secret-from-server {:media-stream-strack (custom-get-media-stream-track-some-how)})

Changed

Demo now includes a checkbox for testing manual provision of a MediaStreamTrack

[2025.11.29.1007] - 2025-11-29

Changed

BREAKING CHANGES

This release migrates to Open AI's GA realtime models. The schemas and payload structures have had significant changes. See Open AI's Beta to GA migration guide.

• malli is no longer a hard dependency
• JSON handling for the JVM and demos switched from cheshire to charred
• schema updated to account for GA realtime models
• JVM and WebRTC transports updated to account for GA
• Demo application now supports toggling content types with output modalities
  • The former :modalities key is now :output_modalities and supports either ["audio"] or ["text"] (but no longer both)