Agentic Web Reference Architecture

A Reference Model for Machine-to-Machine and Human-to-Machine Knowledge

Identifier: NS-AW-001
Version: 1.0
Status: Public Reference Model
Canonical URL: https://www.null-state.dev/agentic-web


Abstract

The Agentic Web requires web resources that can be understood not only by human readers but also by software agents.

An agentic resource should expose identifiable entities, capabilities, relationships, provenance, constraints, and machine-readable representations.

The Null-State Agentic Web Reference Architecture describes these components as a practical reference model.

It is a project-level reference architecture, not an official Internet standard.


1. Core Objects

The model defines the following conceptual objects:

ENTITY

CAPABILITY

ACTION

TOOL

PROTOCOL

RESOURCE

CLAIM

EVIDENCE

PROVENANCE

CONSTRAINT

IDENTITY

VERSION


These objects may be related through explicit identifiers and machine-readable links.


2. Entity

An Entity is an identifiable object represented within a knowledge system.

Examples:

NS:ENTITY:MATHEMATICS

NS:ENTITY:ARTIFICIAL-INTELLIGENCE

NS:ENTITY:AGENT


An entity may represent a concept, organization, software system, dataset, document, protocol, or other identifiable object.


3. Capability

A Capability describes something an agent, service, or system can perform.

Example:

NS:CAPABILITY:SEARCH

NS:CAPABILITY:RETRIEVAL

NS:CAPABILITY:SUMMARIZATION

NS:CAPABILITY:VALIDATION


A capability SHOULD be distinguishable from authorization.

Being technically capable of performing an action does not imply that the action is authorized.


4. Action

An Action represents an operation performed by an agent or system.

Example:

Agent

   ↓

Action

   ↓

Tool

   ↓

Result


An action MAY have:


5. Tool

A Tool is an executable resource that can be invoked by an agent or system.

A tool description SHOULD identify:

name

identifier

input schema

output schema

capabilities

permissions

limitations

version

endpoint

provenance



6. Protocol

A Protocol defines how multiple participants communicate or coordinate.

A protocol reference SHOULD identify:

protocol identifier

version

participants

messages

inputs

outputs

constraints

failure states

security considerations



7. Evidence

An agentic system should distinguish a claim from the evidence supporting the claim.

Example:

CLAIM

  │

  ├── EVIDENCE

  │

  ├── SOURCE

  │

  └── PROVENANCE


This distinction is central to reproducible machine reasoning.


8. Provenance

A provenance record describes where information came from, how it was transformed, and what entity or process produced it.

Null-State recommends compatibility with established provenance approaches rather than inventing an isolated provenance vocabulary.

The W3C PROV family provides an established framework for interoperable provenance information.

Reference:

https://www.w3.org/TR/prov-overview/


9. Machine Representation

A minimal agentic object can be represented as:

{

  "id": "NS:AGENT:EXAMPLE",

  "type": "Agent",

  "capabilities": [

    "NS:CAPABILITY:SEARCH"

  ],

  "version": "1.0",

  "provenance": {

    "source": "https://www.null-state.dev/agentic-web"

  }

}


JSON-LD may additionally be used when linked-data semantics are required.

Reference:

https://www.w3.org/TR/json-ld/


10. Relationship Graph

The basic relationship model is:

ENTITY

  │

  ├── HAS_CAPABILITY → CAPABILITY

  │

  ├── PERFORMS → ACTION

  │

  ├── USES → TOOL

  │

  ├── FOLLOWS → PROTOCOL

  │

  └── PRODUCES → CLAIM

                          │

                          └── SUPPORTED_BY → EVIDENCE

                                                  │

                                                  └── DERIVED_FROM → SOURCE


This representation is intentionally simple enough to be implemented by conventional web applications, APIs, datasets, and AI systems.


11. Interoperability

The purpose of this architecture is not to require one software stack.

A reference object can be implemented using:

The same conceptual identifier can therefore have multiple machine-readable representations.


12. Relationship to Existing Web Standards

Null-State builds on established Web technologies rather than replacing them.

Relevant references include:

W3C JSON-LD 1.1
https://www.w3.org/TR/json-ld/

W3C PROV-O
https://www.w3.org/TR/prov-o/

Schema.org WebPage
https://schema.org/WebPage

Schema.org CreativeWork
https://schema.org/CreativeWork

These technologies and vocabularies provide established mechanisms for structured, linked, and attributable information.


13. Null-State Role

Null-State provides a project-level reference layer for experimenting with these ideas across:

knowledge → protocols → agents → evidence → provenance → machine-readable representations.

The project is intended to make these relationships easier to inspect, reference, reuse, and connect.


14. Status

This is a public Null-State reference architecture.

It is not an official W3C, IETF, ISO, governmental, or academic standard.

External standards remain authoritative within their respective domains.

Canonical reference:
https://www.null-state.dev/agentic-web