Mistral's training default
Mistral's defaults depend on the plan, and one of them is not written down.
- Free mode (Mistral AI Studio): Mistral says it may use your API inputs and outputs to train its models. You can opt out at any time.
- Pay-as-you-go: Mistral's help center says customers "retain full control over this processing and have the right to opt out at any time". It does not say whether training is on or off before you touch the setting.
- Feedback: rating a response with thumbs up or down lets Mistral use that rating, and the input and output, to train its models on every plan.
Mistral's commercial terms (effective 25 September 2026) say it will not use customer data or outputs to train its models except when you opted in on a product that is opt-out by default, or did not opt out on a product that is opt-in by default. Because the API default is not stated, the safe step is to switch training off yourself: in the admin console, open Privacy and disable the "Anonymous improvement data" toggle. Mistral's API toggle and its Vibe (consumer app) toggle are separate, so opting out of one does not opt you out of the other.
When Meetily users select Mistral as their summary provider via BYOK, the request goes to the Mistral API under your API key, and your account's settings apply.
Retention
For the standard APIs, Mistral keeps your Input and Output for the time needed to generate the Output and then for 30 rolling days to monitor abuse, after which it is deleted, unless zero data retention is activated. Two API paths store data longer: the Agents API retains data until you terminate your account, and the Fine-Tuning API retains it until you delete it from Mistral AI Studio or terminate your account.
Mistral's consumer chat app, Vibe (formerly Le Chat), keeps conversations until you delete them or your account, and is out of scope for this page. For the binding text, consult the current privacy policy and DPA on legal.mistral.ai before deployment in regulated environments.
Zero data retention
Zero data retention is available, but it is scoped:
- Plan: pay-as-you-go only.
- Endpoints: stateless API calls only:
/v1/chat/completions,/v1/fim/completions,/v1/embeddings,/v1/moderations,/v1/ocr, and/v1/audio/. Agents, batch processing, the Files API, and Vibe are not covered. - How: submit a request through Mistral's Help Center or support team with your reasons. Mistral approves requests at its discretion.
When ZDR is on, the 30-day abuse-monitoring window does not apply.
Where your data is processed
Mistral is a French company and its DPA is written around GDPR, but EU processing is not the default. For the global endpoint, api.mistral.ai, Mistral says it does not commit to a specific inference location.
Regional inference is available as an option:
api.eu.mistral.aifor the EU, andapi.us.mistral.aifor the US.- Regional endpoints cost 1.1 times standard list pricing.
- They support function calling, but not other tool calls or stateful features (agents, batch, the Files API), and some models may not be offered in every region.
- Account configuration, API keys, billing, and usage data may still be handled outside the chosen region. Only the inference step is regional.
What the Mistral DPA says
Mistral's Data Processing Addendum (effective 27 July 2026) is the contract that governs personal data in your API traffic. Three points matter for meeting transcripts:
- Training: the DPA lets Mistral act as controller to train its models "in accordance with its Privacy Policy, unless Customer is or has opted-out of training". Opting out in the admin console is what removes this.
- Feedback: thumbs up or down feedback, with the associated input and output, is used to train Mistral's models.
- Abuse monitoring: automated moderation, including abuse monitoring on the APIs, applies except when zero data retention is active.
How Meetily uses Mistral
Meetily routes Mistral traffic through your own API key. Your account's contract terms and privacy settings apply. The transcript text is sent over TLS to the Mistral API for summarization, and the response is returned to Meetily and stored locally on your device. Audio is never transmitted to Mistral at any point.
Before you connect Mistral, switch training off in the admin console and, if your organization needs it, request zero data retention. For zero retention by construction, switch to local Ollama and keep the entire pipeline on-device.
References
This page was last checked on 26 September 2026. Always confirm the current terms against Mistral's own linked policy before a regulated deployment.
- "Privacy Policy", Mistral AI. https://legal.mistral.ai/terms/privacy-policy - accessed 26 September 2026 (effective 3 September 2026). Source for the 30-rolling-day abuse-monitoring retention window and the Agents and Fine-Tuning API exceptions.
- "Can I activate Zero Data Retention (ZDR)?", Mistral Help Center. https://help.mistral.ai/en/articles/347612-can-i-activate-zero-data-retention-zdr - accessed 26 September 2026 (updated 12 August 2026). Source for ZDR being pay-as-you-go only, stateless endpoints only, and approved at Mistral's discretion.
- "Do you use my user data to train your Artificial Intelligence models?", Mistral Help Center. https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models - accessed 26 September 2026. Source for the free-mode and pay-as-you-go training wording and the feedback exception.
- "Can I opt out of my input or output data being used for training?", Mistral Help Center. https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training - accessed 26 September 2026. Source for the admin console Privacy toggle and the separate API and Vibe toggles.
- "Commercial Terms of Service", Mistral AI. https://legal.mistral.ai/terms/commercial-terms-of-service - accessed 26 September 2026 (effective 25 September 2026). Source for the section 4.2 training commitment and its exceptions.
- "Regional inference", Mistral AI documentation. https://docs.mistral.ai/inference/regional-inference - accessed 26 September 2026. Source for the global endpoint's lack of a location commitment, the EU and US endpoints, the 1.1x pricing, feature limits, and control-plane data outside the region.
- "Data Processing Addendum", Mistral AI. https://legal.mistral.ai/terms/data-processing-addendum - accessed 26 September 2026 (effective 27 July 2026). Source for the GDPR-based processing terms and deletion procedures.