Feature Request: Native Nextcloud integration using Whisper-compatible API

[erklu]

Description

Enable users of Sunet Drive (Nextcloud-based) to initiate transcriptions via the built-in Whisper integration. This requires the backend to expose an API that mimics the Whisper API structure defined by OpenAI, as supported by e.g. Nextcloud Talk and Text.

---

Expected Behavior

• Backend exposes /v1/audio/transcriptions or equivalent endpoint
• Accepts audio file + model spec + language
• Returns JSON with transcript or full txt version
• Adheres to Whisper-compatible request/response schema
• Respects user authentication and storage isolation (e.g. Nextcloud tokens)

---

Notes

• See: Whisper API schema
• Allows direct transcription from inside Nextcloud (e.g. with Nextcloud Text or OCR plugins)
• Enables single sign-on/flowless use from Drive

---

Priority

Medium — strongly enhances usability for Drive users, aligns with ecosystem standards. Not needed for MVP but should be prioritised after initial launch

[erklu]

In addition to enabling integration with Sunet Drive, exposing a Whisper-compatible API endpoint also opens up future integration opportunities with other services – e.g., LMS, research data portals, video platforms, or automation workflows (e.g., Zapier/Make).

It is important that this API:

• Supports SSO-based authentication (preferably via SWAMID – SAML or OIDC)
• Associates each job with a user and/or group to support accurate usage statistics
• Exposes relevant metadata so that local administrators can track usage in the admin UI in the web service

The goal is to ensure that all transcription jobs – regardless of entry point – follow the same security, accountability, and reporting model.