Add Telnyx WebSocket STT support for Node agents

[robbedgdev]

Feature request

Add Telnyx Speech-to-Text support for LiveKit Agents JS, ideally with an ergonomic API such as STT.withTelnyx(...) or a dedicated @livekit/agents-plugin-telnyx STT plugin.

Motivation

agents-js currently has LLM.withTelnyx(...) in the OpenAI-compatible LLM plugin, but there does not appear to be an equivalent STT integration for Telnyx. Telnyx now provides a standalone real-time STT WebSocket endpoint, which is useful for voice agents that already use Telnyx as a provider and want low-latency streaming transcription in Node agents.

This should not be implemented as a plain OpenAI-compatible transcription wrapper: Telnyx's STT WebSocket protocol uses query-string configuration, Authorization: Bearer ..., binary audio frames, and transcript JSON messages such as { "transcript": "...", "is_final": true, "confidence": 0.98 }.

References

• Telnyx STT WebSocket lifecycle: https://developers.telnyx.com/docs/voice/stt/websocket-streaming
• Telnyx STT WebSocket parameters: https://developers.telnyx.com/docs/voice/stt/websocket-streaming/parameters
• Python Agents already has a Telnyx STT plugin: https://docs.livekit.io/reference/python/livekit/plugins/telnyx/stt.html
• Related JS pattern: plugins/deepgram/src/stt.ts already implements a WebSocket streaming STT plugin in this repo.

Suggested API shape

Either of these would be useful:

import { STT } from '@livekit/agents-plugin-openai';

const stt = STT.withTelnyx({
  apiKey: process.env.TELNYX_API_KEY,
  language: 'en-US',
  transcriptionEngine: 'Telnyx',
  interimResults: true,
});

or:

import * as telnyx from '@livekit/agents-plugin-telnyx';

const stt = new telnyx.STT({
  apiKey: process.env.TELNYX_API_KEY,
  language: 'en-US',
  transcriptionEngine: 'Telnyx',
  interimResults: true,
});

The exact API is flexible; the important part is first-class support for Telnyx's WebSocket STT transport.

Acceptance criteria

• Supports streaming STT over wss://api.telnyx.com/v2/speech-to-text/transcription.
• Reads apiKey from TELNYX_API_KEY when not explicitly provided.
• Supports Telnyx WebSocket parameters such as transcription_engine, model, language, input_format, sample_rate, and interim_results where applicable.
• Streams audio using the Telnyx wire protocol, i.e. binary audio frames rather than OpenAI realtime input_audio_buffer.append JSON/base64 frames.
• Maps Telnyx transcript messages into LiveKit STT events:
  • START_OF_SPEECH
  • INTERIM_TRANSCRIPT
  • FINAL_TRANSCRIPT
  • END_OF_SPEECH
• Handles graceful close/finalization according to Telnyx's engine-specific WebSocket behavior.
• Includes tests for WebSocket URL construction, auth headers, audio frame forwarding, transcript event mapping, and close/error handling.

Notes

The Python Telnyx STT plugin looks like a useful behavioral reference: it connects with Bearer auth, passes STT options as WebSocket query parameters, streams audio, and converts Telnyx { transcript, is_final, confidence } messages into LiveKit STT events. A JS implementation could reuse the existing stt.STT / stt.SpeechStream patterns from the Deepgram or AssemblyAI plugins.