Trust

Security

How ctxstore protects the memory your agents trust it with.

ctxstore holds working memory for AI agents — decisions, project state, sometimes the things a team cares most about. That earns a plain-language accounting of how it's protected, not a row of badges. Questions this page doesn't answer: info@ctxstore.ai.

Encryption in transit

All traffic — the MCP endpoint, the REST API, and this site — is TLS-only. Plain HTTP is redirected before it reaches the application.

Encryption at rest

Stored facts, embeddings, and account data live on AES-256 encrypted volumes. Backups are encrypted with the same standard before they leave the host.

Tenant isolation

Every fact is stamped with your tenant ID at write time, and every query is filtered by it at the storage layer — not in application code that could forget. One tenant's agent can never read, search, or enumerate another tenant's memory.

Authentication

• OAuth 2.1 for clients that support it (Claude Code, Claude Desktop) — no key to handle at all.
• Persistent API keys (emk_…) for everything else, activated by a one-time email code. Same email, same memory, on every device.
• Keys can be rotated from your account page without losing any stored memory.

Secrets never stored in plain text

Agents paste things they shouldn't. ctxstore runs a write-time filter over every fact: API keys, tokens, cloud credentials, private-key blocks, and password assignments are detected by shape and replaced with a labeled redaction marker before the fact is stored. The fact still lands — the secret doesn't. The pattern set is reviewed and extended as new credential formats show up.

Your data is yours

• Exportable at any time — your facts are queryable through the same API your agent uses.
• Never sold, never used to train models, never shared with third parties.
• Deleting a fact deletes it from the live store; encrypted backups age out on a fixed schedule.

Operational practices

• Infrastructure changes ship through code review and CI — no hand-edited production state.
• Production credentials live in a secrets manager, injected at deploy time; they don't sit in config files or environment dumps.
• Staging-first releases: every change soaks on staging before manual promotion to production.