Skip to content

Security Policy

If you find a security vulnerability in MemStrata software or infrastructure, please email security@memstrata.dev.

  • We aim to acknowledge all reports within 24 hours.
  • We target a patch or mitigation within 7 days for critical issues.
  • We will not take legal action against researchers who follow this policy.

PGP public key is available on our GitHub repository.


The following never leaves your machine by design — it’s an architectural constraint, not a policy:

  • All source code and codebase graph data (~/memstrata/index.db)
  • All AI prompts you send and responses you receive
  • All session telemetry and savings metrics (~/memstrata/telemetry.db)
  • Your LLM provider API keys (stored in OS keychain)

The local harness listens on localhost:8080 only — it is not bound to external network interfaces.

The only outbound network call MemStrata makes is to memstrata.dev/lic for license verification. This call transmits:

DataFormat
Subscription IDOpaque UUID, not email
Hardware ID hashHMAC-SHA256, salted per-license
JWT expiryTimestamp only
MemStrata versionSemver string

No source code, prompts, file paths, or API keys are transmitted.

  • Deployed on Cloudflare Workers — no persistent storage of code or prompts
  • Hardware ID hashes are stored salted and cannot be reversed to raw hardware identifiers
  • License verification logs retained 90 days, then purged
  • All communication over TLS 1.3 with HSTS

  • The browser extension requires host_permissions for supported AI provider domains. These permissions are used only to capture chat text and send it to the local harness. Review the extension source on GitHub.
  • The harness proxy does not validate TLS certificates for localhost connections (expected behavior for local services).

SOC 2 Type II is planned for V7 (post-Series A). If you need a security questionnaire completed for a procurement process, email security@memstrata.dev and we’ll provide our architecture documentation and respond to specific controls.


We do not currently run a formal bug bounty program. Researchers who responsibly disclose valid vulnerabilities will be publicly credited (if they wish) in the release notes.