Glossary

Every term used in the docs and dashboard, defined. Skim once before reading the rest.

agent identity: A first-class entity in Anima. Has its own org link, scoped credentials (`ak_*`), email inbox, optional phone number, and audit history. Distinct from the human's account; one human can own many agent identities.
ak_ key: Agent API key. Scoped to a single agent identity. The default credential your agent code authenticates with. Revoked or rotated independently of other agents and the human's master key.
mk_ key: Master API key. Org-scoped admin credential. Required for billing settings, agent creation, hard-cap writes, and anything that affects multiple agents. Held by the human, not the agent runtime.
sk_ key: Service API key. Server-to-server inside Anima's own platform. Not issued to customers. Mentioned in the API reference for completeness.
correlation ID: A workflow-level identifier that threads through every Anima action in one logical task — email reply chains to voice callback to vault read all share one ID. Server-generated for inbound events, propagated through HTTP headers (x-correlation-id) and the SDK. The /audit page renders the full chain.
request ID: Per-HTTP-call unique ID, returned in x-request-id. Distinct from correlation ID. Use request ID for "what happened on this exact call" debugging; use correlation ID for "what happened across this whole workflow."
TCPA: Telephone Consumer Protection Act (US). Caps statutory damages at $500-$1,500 per call placed without consent. Anima requires a `consent_source` assertion on every outbound voice call; the gate runs server-side.
RND: Reassigned Numbers Database (FCC). Tracks phone numbers that have been disconnected and reissued to a new subscriber. Calling under the assumption it's still the original owner is a TCPA violation. Anima checks every outbound voice call against RND via Twilio Lookup, with a 30-day Postgres cache.
consent_source: TCPA gate input. One of: opt-in:web-form, opt-in:double-opt-in, customer-initiated, business-relationship, prior-express-consent. Required on every outbound voice call. Logged with the call record for audit.
DNC: Do-Not-Call Registry (US, FCC). Anima does NOT scrub against DNC server-side — it's the customer's responsibility (CPaaS-standard posture). See /trust/tcpa-dnc.
hard cap: Spend ceiling enforced server-side per meter (email, SMS, voice). Default ON, default $5,000/mo. Warns at 80%, warns again at 90%, blocks at 100%. Per-meter override lets the operator switch a single meter to alert-only mode (notify but don't block) — useful for production agents where pausing mid-customer-call is worse than overspend.
Idempotency-Key: Header on POST/PUT/PATCH/DELETE. We cache the response for 24 hours and replay on retry. Stripe pattern. Constraints: 1-255 ASCII printable. Reuse with a different body → 409 IDEMPOTENCY_BODY_MISMATCH. /api-reference has the full contract.
MCP: Model Context Protocol. Open spec for connecting LLMs to external tools/resources. Anima exposes 190+ tools at https://mcp.useanima.sh/mcp — one URL, path-routed per domain (email, voice, vault, phone, platform, agent).
vault: Per-agent encrypted credential storage backed by bw-serve + Vaultwarden. Every read/write is audit-logged. Plaintext fields never touch disk; the underlying provider sees only ciphertext. Per-credential masking ensures sensitive fields (refresh tokens, etc.) never reach agent-visible responses by default.
OBO: On-Behalf-Of token exchange. Once part of the OAuth-gateway product surface; deprecated 2026-04-25 (no-Composio decision). The vault still stores OAuth credentials; OBO-style scoped-token minting is no longer exposed as a customer-facing feature.
trust score: Anti-abuse signal computed at signup. Three inputs: domain age (RDAP lookup), IP ASN classification (IPinfo Lite or pay tier), Stripe SetupIntent CVC + ZIP checks. Pending signups land in /admin/review with the per-signal verdict. Default OFF (FEATURE_TRUST_SCORE_GATE).
Anima identity sub-types: agent_unverified — restricted mode after sign-up. agent_verified — verified via OTP. claimed — human signed up on console with a payment method on file. Capabilities depend on sub-type and tier.
audit chain: The ordered list of events sharing one correlation ID. Visible at /audit, queryable via GET /v1/audit/events?correlation_id=..., streamable via GET /v1/events/stream. Backed by the audit_events table.