how your assistant
uses AI.

the services we use

Patches is configured to route requests through Concentrate, with Amazon Bedrock as a fallback. Which model handles a request can vary with the configured route.

• Concentrate routes model requests through its API. It receives the conversation and selected context needed for a response. Concentrate’s privacy policy.
• Amazon Bedrock provides hosted model inference through AWS. It receives model requests when that route is selected. Bedrock’s data handling.

what goes into a request

A request can include your messages, relevant conversation history and data returned by permitted abilities. Selected text sources are filtered by the assistant’s access and the conversation’s audience before they enter model context.

Calendar details and source text may contain personal information. Give your assistant access only to information you intend it to use.

what stays outside the conversation

Account credentials and service tokens are resolved by the application when an allowed action runs. They are not supplied as model instructions or tool arguments.

how actions are checked

Model output cannot grant access. The application checks the actor, audience, installed ability and selected resources. Private calendar changes require exact approval and record their outcomes.

AI can misunderstand a request or produce an incorrect answer. Review proposed changes and check the recorded result before relying on them.

your data

We do not train models on your content. Third-party processing is subject to the providers’ terms and our privacy policy. We store conversation and work records to operate the service; this page does not promise zero retention.

Questions about what your assistant can access? Contact us or email [email protected].