AI and Connector Data Policy

Direct user input is the strongest signal in Memrov. If a user corrects a statement through Memrov, contradicting provider-authored statements are kept for audit where appropriate but are not delivered as connector truth.

• Claude memories, ChatGPT bio rows, and similar provider artifacts are treated as what an AI system wrote or stored about the user.
• Memrov may parse, bucket, and display provider-authored statements for review, but the provider remains the author until the user approves, edits, or replaces them.
• Raw conversations and pasted documents are evidence and SQL context first, not default connector context.

• Unresolved conflicting statements may be shown with author labels and relative order when doing so helps the receiving agent understand uncertainty.
• After a user chooses or writes the correct statement, Memrov suppresses conflicting provider statements from connector delivery.
• Conflict handling is not a claim that Memrov discovered truth. It is a governance rule that protects user authority.

• Safe AI is separate from the connector profile. It may route prompts to selected third-party models through configured providers.
• Memrov tracks usage and cost so users cannot accidentally run up excessive spend.
• Where available, Memrov prefers provider settings that minimize retention and training use.

Memrov does not train a Memrov-owned foundation model on private imports. Third-party model providers may have their own processing rules for requests you choose to send through Safe AI or validation flows.