We've released Conversational AI Engine v2.9 on July 1, 2026. This release adds finer control over turn-taking, a lower-cost greeting option, and a new digital human provider.
What's new in v2.9
Manual turn detection
- Clients can now explicitly signal when the user starts or ends speaking, instead of relying on VAD or semantic detection. Useful for interviews, quizzes, and other structured scenarios where false interruptions hurt UX.
Greeting via audio URL
- You can now play a pre-recorded audio file as the greeting instead of generating it via TTS. Falls back to TTS automatically if the file isn't reachable.
Agent state notifications via RTM
- Agent
listening/thinking/speakingstate changes are now also pushed over RTM Message, making it easier to drive UI state and observability. Existing Presence / DataStream flows are unchanged.
Improvements you'll notice
Smarter greeting handling
- When the greeting is non-interruptible, user inputs arriving during the greeting are now merged into a single message before being sent to the LLM, avoiding fragmented replies.
For full technical details, please refer to the release notes