Agent Security Model
Agent-facing access should be explicit, scoped, and observable. This page outlines the model BaseKV is moving toward.
Human owner first
A verified human account remains the owner-of-record for billing, policy, and recovery.
Scoped machine identity
Agents get dedicated service accounts and limited scopes rather than broad admin secrets.
Server-side enforcement
Rate limits and plan quotas must be enforced in API paths, not only in the UI.
Auditability
Every mutating agent action should be traceable to a token, account, and timestamp.
Baseline controls for production usage
- Short-lived or revocable API tokens.
- Per-token and per-account rate limits.
- Idempotency keys for mutation endpoints.
- Budget caps and usage alerts.
- Credential rotation and revocation workflows.