Security
Remembr V1 is designed around scoped access, soft deletion, and operator visibility rather than a flat memory bucket.
Auth model
- Human users authenticate through
/auth/*and receive JWTs. - SDKs and adapters should use API keys from
/api-keys. - Every request carries an effective org scope. Optional team, user, and agent scope narrow it further.
Scoping
- Organization isolation is the baseline control.
- Sessions are created inside the caller's writable scope.
- Search and history queries inherit that scope automatically.
- Structured tags help organization but do not replace security boundaries.
Row-level isolation
Remembr uses database-level scoping to prevent cross-tenant reads. Application code resolves the request scope, and the database session enforces the final isolation boundary.
PII
V1 does not promise automatic PII detection or redaction. Teams should:
- Avoid storing secrets in memory content
- Keep sensitive IDs in metadata only when necessary
- Prefer scoped API keys over shared org-wide keys
- Use deletion workflows for data subject requests
PII automation and policy tooling are forward-looking concerns, not solved by V1 alone.
Encryption
- TLS should terminate at your ingress or platform edge
- Managed Postgres and Redis should use encrypted transport where available
- At-rest encryption depends on your database and volume provider