AI Subprocessors

Last updated: 27 July 2026 · Heurivon Inc.

Heurivon does not operate AI provider accounts on behalf of customers. On customer accounts, all AI processing runs under API keys you supply, on your own account, governed by your agreement with that provider.

A small number of internal accounts — Heurivon staff and a limited group of early beta testers, never customer accounts — run on Heurivon-operated provider accounts. On those accounts, sharing of inputs, outputs, evaluation and fine-tuning data with the provider is disabled at the organization level. Internal users may also supply their own keys, which take precedence over Heurivon's.

OpenAI — required

• Service: Language, Whisper transcription, OCR, image generation, web search
• Purpose: Powers core services including search, OCR, transcription and image generation, which are not model-switchable, and is available for chat and reasoning. An OpenAI key is required to use Heurivon.
• Data handling: OpenAI does not use API data to train its models at any tier. Standard abuse-monitoring retention applies.
• Note: your OpenAI organization has a setting called "Share inputs and outputs with OpenAI," offered in exchange for free daily token allowances. It is disabled by default. If you enable it, your content — including any Google Workspace content — will be used to train OpenAI's models. We recommend leaving it disabled.
• Link: OpenAI Services Agreement

Optional providers for chat and reasoning

Supplying a key for any of the providers below makes that provider selectable for chat and reasoning. Requests are made under your account, and your agreement with that provider governs how your data is handled.

Anthropic

• Service: Language
• Purpose: Selectable for chat and reasoning when you supply a key.
• Data handling: Customer data not used for training under Anthropic's Commercial Terms of Service, at any tier.
• Link: Anthropic Commercial Terms of Service

Mistral AI

• Service: Language
• Purpose: Selectable for chat and reasoning when you supply a key.
• Data handling: Data sent through Mistral's API is not used for model training. Heurivon blocks Mistral Labs models server-side, as those may be used for training regardless of plan or opt-out settings.
• Note: Mistral's Admin Panel includes an organization-level setting permitting API calls to be used for training. It is off by default. If you enable it, your content will be used to train Mistral's models.
• Link: Mistral AI Terms of Service

Groq

• Service: Low-latency inference
• Purpose: Selectable for chat and reasoning when you supply a key.
• Data handling: Customer data not used for training under the Groq Services Agreement.
• Link: Groq Services Agreement

Google Workspace data

Google Workspace content is sent to an AI provider only as needed to fulfil a request you make, and only to providers whose terms exclude customer data from model training irrespective of plan. Heurivon does not use, transfer, or sell Google Workspace data — raw, aggregated, anonymized, or derived — to create, train, or improve any foundational or generalized AI model.

See our Google Workspace Limited Use Disclosure.

Our requirement

Before integrating any AI provider, we verify that their terms prohibit using customer data to train or improve their models. We do not integrate providers that reserve the right to train on customer content.

We will update this page when providers are added or removed.

For questions about our AI data practices, contact privacy@heurivon.com.