Subprocessors

Who helps us run Willow

Last updated October 19, 2026

These are the providers that can process your company's data when Willow runs for you. A provider receives data only when its service is needed for a task or when you turn on the feature that uses it.

Data can include account and contact details, conversations and tasks, company knowledge and files, authorization for connected apps, and usage records.

Which providers apply to you

Dedicated Willow server

Willow runs your workspace on a server reserved for your company, in the US or the EU, and provides the AI models. The providers on this page apply.

Your own infrastructure

Willow runs in your cloud or data center with your own AI model accounts. None of the hosting or AI providers on this page are involved. Composio applies only if you connect apps through it.

Infrastructure

  • Netcup

    The dedicated server that runs your company's isolated workspace, in the US or the EU.

    Data Everything stored in the workspace: conversations, company knowledge, files and settings.

  • Hetzner

    Servers for the Willow control plane and the AI model gateway.

    Data Account and membership data, usage records, model requests passing through the gateway.

  • Cloudflare

    Web delivery and encrypted connections (TLS) at the network edge.

    Data Traffic in transit.

  • Tailscale

    Private network between Willow's own servers.

    Data Connection metadata only. Traffic is encrypted end to end.

AI and search

  • OpenRouter

    Routes requests to AI models and runs web search.

    Data Prompts, tool results and screen images sent for a task, after secrets and financial identifiers are removed.

  • Anthropic, Google, Mistral

    AI models reached through OpenRouter: Claude (served through Amazon Bedrock or Google Vertex), Gemini, and Voxtral for speech to text.

    Data The request for that task only.

  • OpenAI

    AI models reached through OpenRouter. Only when a model alias is mapped to an OpenAI model.

    Data The request for that task only.

  • Vercel AI Gateway

    Small internal decision models that sort and check work, and optional chat models.

    Data Short, redacted task context.

  • Exa

    Search engine behind web search.

    Data The redacted search query only.

Tools and connections

  • Composio

    Connects the apps a company chooses, such as QuickBooks, HubSpot or mail, and runs those actions. Only for apps the company connects.

    Data Authorization tokens and the data of each connected action.

Sign-in and email

  • Resend

    Sends sign-in links and account notifications. Not used for sign-ins through your own Google or Microsoft single sign-on.

    Data Email address and the content of that email.

Zero data retention

  1. I

    Before any text leaves Willow, secrets such as passwords, API keys and tokens, and financial identifiers such as card and bank account numbers, are replaced with a placeholder. Names, email addresses and phone numbers stay, because the work needs them.

  2. II

    Every request through OpenRouter asks for zero data retention and excludes providers that collect data. Requests through Vercel AI Gateway ask for zero data retention as well.

  3. III

    Screen images and audio are not redacted. Screen images are sent only to be described and are not stored. Speech to text uses a model with a zero-data-retention endpoint.

  4. IV

    Zero data retention covers the model request itself. It does not cover what Willow stores in the company's own workspace, or apps the company connects.

  5. V

    Company data is never used to train AI models.

Companies that bring their own AI model accounts are covered by their own agreement with that provider.

Changes and questions

When a provider is added or replaced, this page changes and the date at the top moves. Customers with a data processing agreement get notice and the right to object as set out in that agreement.

Questions about a provider or how it handles data: support@willowai.app