Skip to content

Security: AndrewMoryakov/AgentMemory

Security

SECURITY.md

Security Policy

Scope

AgentMemory is currently a local-first public alpha project.

Memory content may include sensitive project, user, or agent data. Treat local runtime exposure and shared machine access accordingly.

Reporting A Security Issue

Please do not open public issues for sensitive security reports.

Instead:

  • contact the maintainer privately if possible
  • include reproduction details, affected surface, and impact
  • state whether the issue affects local-only use, Docker deployment, or MCP/HTTP exposure

Current Security Posture

  • local-first by design
  • not positioned as a hosted multi-tenant service
  • auth and remote exposure should be treated cautiously
  • user_id is taken from the request payload by default, so any valid credential can name any scope. AGENTMEMORY_ENFORCE_AUTH_USER_ID=1 binds the scope to the identity the credential was issued for — see docs/AUTH_IDENTITY_BINDING.md, including what that mode does not guarantee

Out Of Scope For Public Claims

Until documented otherwise, the project should not be described as:

  • hardened for hostile multi-tenant environments
  • production-ready for open network exposure
  • a managed hosted security boundary

There aren't any published security advisories