Skribby only supports transcription providers that do not train on Skribby customer audio or transcripts.

For provider-side retention, Skribby uses the strongest retention control available for each provider:

- Zero Data Retention (ZDR) or zero-retention mode.
- Deletion after processing when the provider creates a stored job or artifact.
- Realtime streaming paths where audio is processed ephemerally instead of stored as a transcription job.

This page covers transcription-provider processing. Skribby's own recording and transcript retention is controlled separately by bot settings, workspace retention settings, storage add-ons, and delete requests.

## Standard Providers

These providers meet Skribby's no-training and low-retention requirements without a provider-specific caveat:

- Groq
- Soniox
- AssemblyAI
- Deepgram
- Speechmatics
- Rev AI
- Gladia
- Salad
- Mistral/Voxtral
- OpenAI
- `none`

## Realtime Transcription

Realtime transcription is part of Skribby's standard privacy posture. For realtime models, ephemeral streaming is the intended low-retention path: audio is processed live instead of being stored as a batch transcription job.

Realtime providers covered by this posture include:

- Soniox realtime
- AssemblyAI streaming
- Deepgram realtime
- Speechmatics realtime
- Gladia realtime

## Important Caveats

Most supported providers meet the standard posture above. The following providers have additional retention limitations:

- **ElevenLabs**: no-training is in place, but ElevenLabs transcription does not currently have a standard ZDR or deletion path.
- **xAI**: xAI speech-to-text retains API inputs and outputs for up to 30 days before automatic deletion. Treat it as low-retention with automatic deletion, not strict ZDR or immediate deletion after use.

For strict zero-retention requirements, choose one of the standard providers above or ask Skribby to restrict provider availability for your workspace.