https://livekit.io logo
Join Slack
Powered by
Can Adaptive Interruption Handling be used in self...
# ask-ai
b
Can Adaptive Interruption Handling be used in self-hosted agents (Typescript)?
✅ 1
t
Thanks for reaching out, Alejo. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — you’re asking whether Adaptive Interruption Handling can be used in self-hosted agents built with TypeScript. Short answer: yes. What this means and what to check • Adaptive Interruption Handling is available for self-hosted agents (including TypeScript/Node.js). • It’s enabled by default for agents running in LiveKit Cloud or in dev mode when you’re using TypeScript SDK v1.2.0+ and meet the other requirements below. • You need an STT model that supports aligned transcripts (check 
capabilities.alignedTranscript
for your STT). Adaptive handling relies on aligned transcripts to work correctly. • The adaptive model runs on LiveKit Cloud’s inference infrastructure even for self-hosted agents, and you get a shared quota of 40,000 inference requests per month across all plans (suitable for development and experimentation). How to enable it in TypeScript 1. In your agent session configuration, set 
interruption.mode
to 
"adaptive"
inside 
turnHandling
. Example:
Copy code
const session = new voice.AgentSession({
  // stt, llm, tts, vad, etc.
  turnHandling: {
    turnDetection: new livekit.turnDetector.MultilingualModel(),
    interruption: {
      mode: 'adaptive',
    },
  },
});
If you prefer VAD-only detection, set 
interruption.mode
to 
"vad"
. If you want, tell me: • Which STT model you’re using (so I can confirm aligned transcript support), and • Your SDK version (to confirm you have v1.2.0+), and I’ll help verify your setup or show the exact configuration snippet for your codebase. Sources: Adaptive interruption handling | LiveKit Documentation | Solving unwanted interruptions with Adaptive Interruption Handling | LiveKi..., with interactive elements
b
in the docs it says that only 40000 inference requests are allowed for self hosting
t
Thanks — I’m following up on your point about the 40,000 inference-request limit for self-hosted agents. I’ll clarify what that limit means and how it may affect your setup. What the 40,000 requests/month limit means • The 40,000 inference requests is a shared monthly quota for Adaptive Interruption Handling when used with self-hosted agents (including TypeScript). • That quota is applied across all your usage of the adaptive interruption inference service (not per agent instance). Practical impact • This quota is usually sufficient for local development, testing, and early experimentation. • If you expect production traffic with many concurrent sessions or high call volume, 40k requests/month can be reached quickly and may limit adaptive interruption calls. Technical notes • Even for self-hosted agents, the adaptive interruption model runs on LiveKit Cloud’s inference infrastructure (so those inference calls count against the quota). • If your STT supports aligned transcripts and you enable 
interruption.mode = 'adaptive'
, those adaptive inference requests will be generated automatically as part of turn handling. Options and next steps • If you think 40k/month is enough, go ahead and enable adaptive mode as shown earlier; I can help verify your config. • If you anticipate higher volume, tell me roughly how many simultaneous users or sessions and expected calls per session (or how many minutes of audio per month). With that info I can help estimate monthly inference usage and suggest alternatives (for example, sticking with VAD-only detection by setting 
interruption.mode = 'vad'
to avoid adaptive inference calls). Would you like help estimating your expected usage or a walkthrough to switch to VAD-only if the quota is a concern? , with interactive elements
3 Views
Open in Slack
PreviousNext
Powered by