@assemblyline-agents/audio
v4.0.0
Published
Pre-model audio processing adapters for Assembly Line agents.
Maintainers
Readme
@assemblyline-agents/audio
Framework-level pre-model audio processing for Assembly Line agents.
import { defineGateway } from "@assemblyline-agents/core";
import { openRouterAudioTranscription } from "@assemblyline-agents/audio";
export default defineGateway({
media: openRouterAudioTranscription()
});Set OPENROUTER_API_KEY. The default model is
openai/whisper-large-v3-turbo; override it with
OPENROUTER_AUDIO_TRANSCRIPTION_MODEL or gateway options. Successful
transcripts are stored in the configured private blob adapter and reused by
attachment hash. Transcript text is supplied to the model as untrusted derived
context and is not copied into run audit events.
Apple CAF voice notes are converted to M4A with the package's bounded FFmpeg binary before they are sent to OpenRouter. The original private attachment is not modified. M4A and the other provider-supported formats are sent unchanged.
