Build agents that listen, respond and handle interruptions.
Connect WebRTC rooms, live speech recognition and synthesis to your LLM using dTelecom SDKs and prepaid gateway sessions.
One session bundle
Create WebRTC, STT and TTS together with x402-client. Pass the returned tokens to agents-js and the browser client.
See words early
Show provisional transcript revisions in chat, then replace them with the authoritative final. Partials can interrupt playback; finals start LLM turns.
Control speech
Independently enable Smart Turn, choose Kokoro/Supertonic voices and speed, or explicitly enable watermark. Use complete SSML for greetings and announcements; ordinary streamed LLM replies remain plain text.
Usage example
Two audio participants, one STT session and 367 spoken TTS characters/minute cost approximately $0.0064/minute, excluding LLM charges. This is an example, not a flat bundle rate. See all billing units.
SDK responsibilities
agents-js handles the live voice pipeline. x402-client manages credits, sessions and separate translation, file/WAV, vision and moderation requests. These additional APIs are not automatically invoked by the agent. Applications own moderation sampling and actions.
Try it
AI language tutor demonstrates multilingual speech and provisional chat transcripts. See the demo source and its pinned SDK versions.