Flux, the speech recognition company, has updated its v2 WebSocket API to allow developers to change configuration settings during a live audio stream, removing the need to disconnect and reconnect when conversation context shifts.
The new Configure message lets developers update keyterms, end-of-turn (EOT) thresholds and timeout settings on an open stream, with changes taking effect immediately on all subsequent audio.
Previously, teams had to choose between loading every possible keyword at connection time, which created unwieldy lists, or keeping lists short and accepting lower accuracy for task-critical phrases; changing configuration mid-call required closing and reopening streams or managing multiple simultaneous connections.
The update is aimed at voice agents that move between distinct tasks during a single call, such as switching from general conversation to an authentication flow, where different keywords and turn-detection sensitivities are needed.
Flux said the feature also allows applications and orchestrators to react programmatically, as each Configure request returns either a ConfigureSuccess or ConfigureFailure response.
Related reading
- Google brings Find Hub luggage tracking to airlines
- OpenAI publishes GPT-5.3 Instant system card
- Gemini 3.1 Flash-Lite enters preview
"No ceremony," the company said in its announcement, describing the simplicity of sending a single message on the existing WebSocket connection.
The feature is available now in the Flux v2 WebSocket API, with documentation and a quickstart guide published alongside the announcement.
The recap
- Flux adds a Configure message to its v2 WebSocket API
- Key fields include keyterms, eot_threshold and eot_timeout_ms settings
- On-the-fly configuration is available now in Flux v2 API