AI & assistants Available
Mistral (Voxtral)
A meeting wants to know who said what, not just what was said. A provider that decides that per request and takes only a quarter of an hour per request numbers the speakers afresh for every part. Voxtral takes the whole meeting in one go.
What you need
- A Mistral account with an API key. You settle with Mistral directly.
- The ai.settings.manage permission to configure the provider. Admin-only by default.
- A feature that uses the microphone switched on: meeting transcripts and minutes, Urenhulp, task dictation or the Assistent
- The app reachable over https, or the browser grants no microphone access
How to connect it
- 1 Create an API key for this installation in your Mistral console.
- 2 Open Instellingen → AI (Settings → AI) and go to the Spraak naar tekst fieldset.
- 3 Set Spraakaanbieder (speech provider) to Mistral (Voxtral) and fill in API-sleutel voor spraak. It is stored encrypted and never played back.
- 4 Spraakmodel may stay empty:
voxtral-mini-latestis used then. There is no base URL to fill in. - 5 Under Functies, switch on where the microphone should appear, for example Vergadertranscripten en verslagen, and optionally set a Maandbudget audio (seconden).
- 6 Press Verbinding testen (test connection) and grant
ai.usein Instellingen → Rollen to the roles allowed to use the features.
Why a separate speech provider
Anthropic, the default provider for the chat features, has no transcription service. So speech always has its own setting with its own key, and reusing the chat key configures nothing. Mistral is a choice of its own there because Voxtral's transcription API can be asked two things a plain OpenAI-compatible call cannot send: who spoke, and when. Mistral's chat API you use, if you want to, through the OpenAI-compatible provider.
A meeting in one request
A provider labels speakers per request. If a recording has to go in parts, the speakers in the second part are called S1 and S2 again, so schakl. numbers them on rather than guessing who was who. With Voxtral that is unnecessary up to three hours: one request, one set of labels, which you pair with the participants in the review step. The minutes, decisions and action items then come from your chat provider, and only become a contact moment and tasks once somebody confirms them.
Permissions
Grant these to the role that manages this integration (Settings → Roles). Admin-only by default, and never to the client role.
-
ai.settings.manageSet the speech provider and its key. Admin only by default. -
ai.useUse the AI features, dictation and transcription included.
What it deliberately does not do
- Speech to text only. For the assistant, the writing assist and the other features you configure a chat provider.
- Beyond three hours a recording is cut into parts with Voxtral too, and the provider numbers the speakers afresh per part.
- The audio goes to Mistral and nowhere else; usage is counted in seconds, never as the text of a transcript.
- Supported audio containers: WebM, Ogg, MP4/M4A, WAV and MP3, identified by the file's own bytes.
Where to find it Instellingen → AI (Settings → AI), the Spraak naar tekst fieldset, Spraakaanbieder set to Mistral (Voxtral). The record button for a meeting sits under Meetings.
Read the guide