Work with files and audio

Overview

File storage and transcription are separate services. A temporary upload is not durable knowledge and should have an explicit lifetime. Transcription turns audio into text that the existing Operations Desk pipeline can process; it does not require a second conversational architecture.

An upload is a resource with a lifetime

The file service stores content according to its configured provider and expiry policy. A returned file reference is not the same thing as text the agent has read, and it does not automatically ingest the content into GraphRAG. Decide which code downloads, parses or transcribes the content and what bounded result becomes part of the request.

Transcription joins the existing message flow

A transcription service consumes audio and returns text. That text can be shown for review and sent through the same assistant pipeline as typed input. For large recordings, stream/process chunks through the supported interface rather than allocating the entire recording in a UI component. Preserve the association between the transcript, user and source recording.

Turn an attachment into useful input

Download the Operations Desk source. The README lists project setup, package prerequisites and local ports.

1. Use the Host file API with principal ownership and configure FabrCore:FileStorage path, TTL and cleanup. Use an appropriate persistent/shared store when deployed across hosts.
2. Add an optional Azure OpenAI audio client in a plugin/service, stream the upload and pass the transcript as the next request. Keep endpoints and keys outside model arguments.
3. Bound file sizes, split long recordings according to the currently supported model limits, propagate cancellation and keep source provenance where needed.

using var audioStream = File.OpenRead("/path/to/audio.mp3");

var options = new AudioTranscriptionOptions
{
    Filename = "audio.mp3",
    ResponseFormat = AudioTranscriptionFormat.Verbose,
    Temperature = 0f,
    Language = "en"
};

AudioTranscription transcription = await audioClient.TranscribeAudioAsync(
    audioStream, "audio.mp3", options);

Console.WriteLine(transcription.Text);
Console.WriteLine($"Duration: {transcription.Duration}");
Console.WriteLine($"Language: {transcription.Language}");

Go deeper

Explore the related documentation.

• FabrCore Websocket
• FabrCore Client
• FabrCore Blazor
• From a stream (recommended for chunked processing)