xAI

xAI is a speech-to-text plugin. It streams transcripts with support for interim results, speaker diarization, and multichannel audio.

Setup

Set your xAI API key in the worker environment. Generate a key from the xAI site:

export XAI_API_KEY=<key>

Import the plugin and pass it to the pipeline’s stt slot.

from zeroruntime.plugins import XAISTT

stt = XAISTT()

# Pipeline(stt=stt, ...)

import { XAISTT } from '@zeroruntime/js-sdk/plugins';

const stt = XAISTT();

// Pipeline({ stt, ... })

Constructor parameters for XAISTT.

Parameter	Type	Default	Description
`api_key`	`str`	`None`	xAI API key. Falls back to the `XAI_API_KEY` environment variable when unset.
`sample_rate`	`int \| None`	`None`	Sample rate of incoming audio in Hz: 8000, 16000, 22050, 24000, 44100 or 48000. Leave unset to detect it automatically.
`encoding`	`Literal['pcm', 'mulaw', 'alaw']`	`"pcm"`	Encoding of incoming audio: “pcm” (16-bit linear), “mulaw” or “alaw”.
`interim_results`	`bool`	`True`	Emit partial transcripts while the user is still speaking, before the final result.
`endpointing`	`int`	`50`	Milliseconds of silence before an utterance is finalized, 0 to 5000. Keep it low for fast turn-taking.
`language`	`str \| None`	`"en"`	BCP-47 language code, e.g. “en” or “fr”. Any supported language is transcribed; this enables written-form numbers and currencies.
`diarize`	`bool`	`False`	Label each transcribed word with the speaker who said it.
`multichannel`	`bool`	`False`	Transcribe each audio channel separately instead of mixing down to mono.
`channels`	`int`	`2`	Number of audio channels to transcribe when multichannel is on, 2 to 8. Ignored otherwise.
`base_url`	`str`	`"wss://api.x.ai/v1/stt"`	WebSocket URL of the xAI speech-to-text endpoint.
`model`	`str \| None`	`None`	Transcription model, e.g. “grok-voice-transcribe-1.0”. Leave unset to use xAI’s standard model. Python only.

SDK	Import	Constructor
Python	`from zeroruntime.plugins import XAISTT`	`XAISTT()`
Node JS	`import { XAISTT } from '@zeroruntime/js-sdk/plugins'`	`XAISTT()`

Transcribed text passes to the LLM once turn detection decides the caller has finished speaking.