transcribeAudio (plugins/transcribe.ts)

Multi-provider speech-to-text module supporting OpenAI, Deepgram, Gemini, and ElevenLabs with size limits, timeouts, and abort signals.

// Source: https://github.com/cosmicstack-labs/mercury-agent | License: MIT | Copyright 2026 Cosmic Stack
// Provided via justclaw.it.com; keep this notice with the code.
/**
 * Speech to text, with a service the owner already set up in Mercury: the
 * OpenAI provider (gpt-4o-mini-transcribe, falling back to whisper-1), or a
 * Gemini, Deepgram (nova-3) or ElevenLabs (scribe_v1) key saved for voice
 * (Settings → Voice / the Optimus plugin). The voice the owner chose there
 * decides which one (see pickSpeechEngine). Used for voice notes from
 * channels and for dictation in the apps.
 */
import { readFileSync, statSync } from 'node:fs';
import { basename } from 'node:path';

const MAX_BYTES = 25 * 1024 * 1024;

export interface TranscribeCreds {
  apiKey: string;
  baseUrl?: string;
}

ex

... (truncated -- full source via MCP)