Resources · Glossary · Plain-English terms

Speak fluent trust layer.

Plain definitions for the terms we use across the platform: agent identity, non-human identity, policy boundaries, blast radius, attestation, and more. No marketing fluff — just what each thing actually means and where it shows up in your stack.

Identity

Who can act on your stack

Agent Identity

A verifiable identity issued to an AI agent (LLM-driven worker), separate from any human or service account. Scope, rotation, and audit are tied to this identity, not to a shared API key.

Non-Human Identity (NHI)

Umbrella term for any caller that isn't a person: services, scheduled jobs, agents, CI runners. The most-stolen credentials in 2025 belong to NHIs that nobody is rotating.

Credential Scope

The narrow set of actions a credential is allowed to perform. Compass shows current scope per identity; Vault enforces it at issue and call time.

Policy

What they can do

Policy Boundary

The runtime envelope every action is checked against. Cross it and you get a signed deny event — visible in Compass, evidenced in Compli.

Blast Radius

Everything an identity can reach if its credential leaks. The shorter the path from a stolen token to crown-jewel data, the worse your day.

Drift

An identity's actual behavior diverging from its policy envelope — usually slowly, sometimes suddenly. Detected by Compass and Beacon.

Proof

How you know it happened

Attestation

A signed statement that a workload, action, or artifact matches its declared identity. STACK Lattice produces them per running pod; Compli aggregates them as evidence.

Audit Trail

The cryptographically signed log of every decision Vault made: who asked, what scope, was it allowed, what happened. The basis for both Compli evidence and incident replay.

Lineage

The chain that connects a final output back to every input that influenced it. Critical for model outputs (STACK Lineage) and for compliance assertions about how a decision was reached.

Missing a term

Tell us what to add

If your team has a phrase we should define here, send it. We add terms based on what real teams ask about during evaluations.