Skip to content
AX Framework

Getting Started

Core Concepts

Capabilities, resources, actions, grants, receipts, and the agent interaction lifecycle.

AX Framework uses a small vocabulary consistently across the principles, patterns, and docs. This page defines each term.

Agent

Software that pursues a goal on someone's behalf by deciding, at runtime, which actions to take. In these docs, "agent" means an AI agent acting for a person or organization. An agent may operate a browser, call APIs, or use tools exposed through a protocol such as MCP.

Principal

The person or organization an agent acts for. The principal delegates authority and remains accountable for the outcome. Good AX keeps the principal informed and in control.

Capability

Something the product can do for an agent, described in a machine-readable way: a name, a purpose, inputs, outputs, risk, and authentication requirements. Capabilities are what an agent discovers.

Resource

A thing the product holds that an agent can read: a product, an order, a booking, a document, a ticket. Resources have stable identifiers and structured descriptions.

Action

A capability that changes something. Actions are defined as contracts: typed input, preconditions, side effects, and a result. Read-only capabilities (search, list, get) are sometimes called queries.

Workflow

A multi-step task modelled as a resource with named steps, status, and a next action. Workflows let an agent resume after interruption.

Grant

A scoped, time-limited, revocable permission that lets a specific agent perform specific capabilities for a specific principal. Grants replace sharing a person's credentials.

Approval

A pause in an action until the principal confirms it through a channel the product controls. Approvals are bound to exact parameters.

Receipt

A durable record of what an action did: inputs applied, resources affected, amounts charged, the actor and grant, and a timestamp.

Capability manifest

A document listing a product's capabilities, published at a predictable location. See Capability Discovery.

The lifecycle

These concepts meet in the agent interaction lifecycle:

  1. 01

    Discover

    What can I do here?

    Discoverability

  2. 02

    Understand

    What does this mean?

    Comprehension

  3. 03

    Authorize

    Am I allowed?

    Trust & Control

  4. 04

    Act

    Do it, predictably.

    Actionability

  5. 05

    Verify

    Did it work?

    Feedback

  6. 06

    Recover

    What now?

    Recovery

Recover returns to Understand (clarify), Authorize (request scope), or Act (retry, roll back).
The agent interaction lifecycle: Discover, Understand, Authorize, Act, Verify, and Recover, which loops back.
StageThe agent asksConcepts involved
DiscoverWhat can I do here?Capability, manifest
UnderstandWhat does this mean?Resource, state
AuthorizeAm I allowed?Grant, approval, principal
ActDo it.Action, workflow
VerifyDid it work?Receipt, status
RecoverWhat now?Errors, retries, rollback

Last updated 2026-10-06. AX Framework documentation describes proposals unless a section is marked as an established standard.