Information Architecture for Intelligent Agents
Every resource should explain itself. What it is, what supports it, what may happen next, and who has authority.
Whether published by a person, generated by an agent, or changed by an application, IA² keeps identity, relationships, evidence, constraints, and possible actions explicit and inspectable.
The contract promises notice within 72 hours. Internal policy requires 24.
Vendor review example
When an agent says “conditional,” follow every edge.
A security review is more than text to summarize. It contains requirements, claims, evidence, policy constraints, a decision, and an authority boundary.
IA² keeps those concerns distinct and connected in the same interface a person uses. The result can be inspected locally by ordinary software, without a detached duplicate graph.
- Governing policy Notify customers within 24 hours A machine-checkable requirement, not a phrase recovered from a summary.
- Vendor evidence Section 8.2 commits to 72 hours The claim remains linked to the contract that supports it.
- Recommendation Proceed only after amendment The recommendation exposes the policy gap and its rationale.
- Authority boundary The agent drafts. Maya approves. Describing a possible action never grants permission to execute it.
The review graph exposes its runtime decision state, provenance links, and validation contract.
Open the vendor security review Inspect the requirements, evidence, policy gap, decision, and validation contract.A shared layer, not a new silo.
Documents, interfaces, APIs, and agent tools contribute different capabilities. These layers connect them without collapsing data, presentation, and execution into one mechanism.
-
Surfaces and live state
Where is information encountered, and what is changing now?
Documents, interfaces, ActivityStreams, and host-native bindings. -
Identity and relationships
What is this, and how does it connect across surfaces?
RDF 1.2, Schema.org and other open vocabularies, PROF profiles, and R2RML mappings. -
Integrity and evidence
What supports it, and what must remain true?
SHACL, PROV-O provenance, typed confidence, scoped statements, validation, and signatures. -
Interaction and authority
What may happen next, under which conditions, and by whose authority?
Traceable decisions and conversational interaction, with controlled execution. -
Continuity and exchange
How does context survive across surfaces, sessions, and organizations?
DCAT discovery, processing lineage, governed artifact evolution, snapshots, diffs, and subscriptions.
Use the substrate. Inspect the proof.
- HTML/RDF core model An attribute binding for complete RDF 1.2 datasets. Specification
- RDF/HTML A DOM vocabulary and inert renderer for HTML documents described in RDF. Specification and package
- RDF/HTML URL renderer Fetch a Turtle or TriG document and inspect its rendered HTML in an isolated frame. Interactive tool
- Discovery and Enrichment Purposeful discovery plus a portable vocabulary for candidates, contributions, views, roles, and lifecycle state. Specification
- HTML Agent Resource Envelope A browser-openable HTML/RDF envelope for exact, related agent resources. Specification
- Using IA² with HTML Publish RDF in HTML or derive an inspectable HTML document from RDF. Guide
- Live semantic workspace Work, conversation, decisions, evidence, and constraints in one live graph. Interactive demo
- HTML RDF Navigator Extraction, navigation, and serialization tied back to DOM carriers. TypeScript package
- HARE Viewer Optional verified file browsing for manifest-only and authored HARE documents. TypeScript package
The web should not make every agent start from text.
IA² explores a web where meaning, evidence, constraints, decisions, and possible actions travel with the information itself. Any agent can inspect that context, people can see where it came from, and no single platform has to own the shared understanding.
Follow the work on GitHub