Skip to main content
Available under X-Mnemom-Version: 2026-04-15 and later. See Agent cards for the two-card model.

​
Two cards per agent

Every agent on Mnemom is governed by two YAML cards: Both cards are composed at storage time from three scopes — Platform ⇒ Org ⇒ Agent — into a canonical document the gateway and observer read on every request. Composition semantics (union / deny-overrides / floor / concat) are defined on the Agent cards page.

​
Endpoint surface

​
Per-agent

​
Per-org

​
Content negotiation

All card endpoints are YAML-first. PUT endpoints accept either YAML (via Content-Type: text/yaml) or JSON (via Content-Type: application/json with the raw card as the body, or wrapped as { "card_yaml": "<yaml string>" } or { "card_json": {...} }).

​
Required request headers on mutations

Every PUT, POST, or DELETE must carry:
Replays of the same key with the same request body return the cached response. Replays with a different body return 422 Idempotency-Key conflict.

​
Cross-cutting contracts

Every response from a card endpoint includes X-Mnemom-Schema, every mutation writes a governance_audit_log row before responding, and every mutation emits a webhook event. See the Governance guarantees page for details.

​
What is NOT available through these endpoints

  • Trust edges, team cards, team coherence, fault lines — separate endpoints under /v1/teams/…; unchanged by the unified-cards consolidation.
  • Safe House operational endpoints (quarantine, canary, patterns, metrics, sessions, compliance reports) — preserved; unchanged by the unified-cards consolidation.
  • Policy evaluation — /v1/policies/evaluate and /v1/policies/evaluate/historical are preserved as a distinct concern (evaluate a card’s enforcement section against a tool manifest).