Agent-native app glossary
This agent-native app glossary is the Agent-Native Application Vocabulary: the terms the specification is coining, each with a definition precise enough to cite on its own. Start from the core term. An agent-native application is a portable, inspectable, user-owned package that gives a capable AI model its identity, judgment, governance, workflows, and durable state.
Last updated: 2026-08-07
An agent-native application is a portable, inspectable, user-owned package that gives a capable AI model its identity, judgment, governance, workflows, and durable state.
Host
The AI system an agent-native app mounts into: model, inference, tool execution, permission boundary, and runtime surface.
Full definitionMount
Loading a published agent-native app into a compatible Host, gated by an entitlement check and the constitution hash.
Full definitionConstitution hash
A cryptographic fingerprint of an app's governing rules, anchoring each version immutably.
Full definitionPDP (policy decision point)
An independent, model-backed evaluator of governed actions; separate from the agent because an agent can't audit itself.
Full definitionSigned entitlement
A signed grant of the right to mount and run a specific app version, checked by the Host at mount time.
Full definitionAgent card
The package manifest: identity, declared tools and bindings, data custody, compatibility, version, and provenance.
Full definitionDeclared data custody
The package's stated, verifiable answer to who holds which data; users keep theirs by design intent.
Full definitionCapability inheritance
A mounted app improves when the user's Host model improves; model identity is logged because changes cut both ways.
Full definition