Security

AgentGuard is built as an awareness tool for sensitive information. Trust and security are core to its design.

Security model

• Local-first processing
• No external network calls for content analysis
• No backend storage of user data
• User retains full decision authority

Data handling

AgentGuard does not transmit, store, or log document content. Any analysis performed is ephemeral and scoped to the active user session.

Verifiability

AgentGuard's security posture is verifiable by design. Deterministic behavior means identical inputs produce identical outputs, making the system auditable and predictable.

There are no server-side components or external dependencies for content analysis. The entire processing model runs locally and can be inspected.

Signal interpretation

Distribution markers such as "Confidential" or "Do not distribute" are treated as sharing restrictions rather than indicators of content sensitivity. This distinction helps avoid overstating what isolated document labels convey. All signal interpretation is deterministic and local-first — repeatable, auditable, and performed entirely on your device.

Responsible disclosure

If you believe you have discovered a security issue, please report it responsibly: [email protected]