Platform
Nine capabilities.
One substrate.
Each offering is independently adoptable and depends only on the ones below it. Together they are every capability an agentic application needs — from a person's first passkey to the receipt for an AI agent's last action.
01Where it runs
An estate: one deployment of the substrate, many applications around it.
Home for people and ceremonies, a runtime for agents, vaults for records, an edge for admission, discovery and a skills registry, and a chain for the anchor. Your application is a relying app: an OIDC client and an A2A caller.
02Live
The estate, as its people see it.
Screens from the running faithnet estate, signed in as the Home's demo people. Every product here is a relying application of the same substrate.




01 Identity
One canonical agent per person, organization and service.
Smart Agent accounts with passkey control, credential recovery, typed on-chain names and signed agent cards. The address is canonical; everything else points at it.
What you would otherwise stitch
- — Auth0 / Okta / Cognito
- — wallet-connect + a users table
- — a DID method of your own
- — a naming service
Standards
03 Harness
Ask → Intent → Mandate → Plan → Verify → Receipt.
An authority-aware agent loop: the planner proposes, the mandate authorizes, the executor acts, the receipt proves. Durable runs, streamed progress, parallel read steps, schedule and message triggers, per-run bills.
What you would otherwise stitch
- — LangGraph / MAF / CrewAI wiring
- — a human-in-the-loop click UI
- — Temporal for approvals
Standards
04 Edge
Admission at the boundary — HTTPS required, mTLS optional, authority always.
A2A over HTTPS with application authentication, canonical identity resolution and an admission stage that no transport evidence can skip. Private MCP behind admitted runtimes; public MCP only as labelled interop.
What you would otherwise stitch
- — API gateways with bespoke auth plugins
- — service-mesh identity as authorization
Standards
05 Registry Kit
Build your own registry; be what registries are built from.
SA-anchored registry contracts with pluggable admission hooks, admission receipts, a lifecycle log, signed agent cards, discovery projections (A2A card, ARD, ACP registry) and a public knowledge base projected only from chain state.
What you would otherwise stitch
- — a hosted agent directory you rent
- — ERC-8004 / ANS / DNS bridges you maintain
Standards
06 Evidence
Receipts, provenance and audit that outlive the platform.
Verification receipts bound to intent + mandate + step + playbook digest; PROV-O graphs with a trace-context spine; append-only audit with forensics tooling; verifiable credentials for claims.
What you would otherwise stitch
- — LangSmith / Langfuse as the record
- — application audit tables
- — a credentials vendor
Standards
07 Coordination
Work between agents: endeavors, plans, commitments, decisions.
A durable Endeavor with a shared plan, participants found through discovery at the capability level, signed contribution commitments against an exact plan revision, decisions and rules. Never confused with in-agent orchestration.
What you would otherwise stitch
- — a project tool bolted to a chat tool
- — workflow engines as the source of truth
Standards
08 Ontology
How the domain is shaped — bound by IRI, checked at build.
An upper ontology for everything agentic — agents, credentials, custody, delegation, capabilities, situations, provenance — that a domain ontology imports and subclasses. From the two, an agent archetype bundles skills, capabilities and ontology areas; from the archetype, the A2A signed agent card and the SKILL.md packages are projected. Vault records are ontology-shaped; behaviour is generated from the model; a gate fails the build when code invents a term.
What you would otherwise stitch
- — prompt-resident domain rules
- — app tables that disagree with each other
- — an agent card typed by hand
Standards
Ontology, live
Define the domain once. Every agent surface is projected from it.
An upper ontology says what every agentic system has. A domain ontology imports it and says what this domain has. An archetype bundles the skills, capabilities and ontology areas one kind of agent needs. From that, the A2A signed agent card and the SKILL.md packages are generated — never typed by hand. Meaning flows down the ladder; authority never does.




What this replaces
A system prompt that explains the domain, a table in the app that disagrees with it, an agent card typed by hand and a playbook that names things the model has never heard of.
What is checked
SHACL shapes on every class. A SKILL.md whose knowledge.requires names a class the ontology does not have fails to compile. Code that invents a term fails the build.
What is not authority
Any of it. The archetype, the card and the playbook are consulted by no verifier. The coach reads Alice's hands under a Study Grant she signed at her Home — and loses them the moment she revokes it.
09 Operations
Run it: KMS delegates, deploy recipes, live gates, conformance.
Service keys as revocable delegates (never the identity), budget-routed model selection, live verification gates that run nightly against the real estate, `ap conform` for A2A and MCP.
What you would otherwise stitch
- — a secrets sprawl you audit by hand
- — a QA suite that only runs on mocks
Standards
How they compose
Dependencies point one way. No back-edges.
Ring 0 is packages and contracts. Products — a Home, a discovery service, a naming service, your application — live in their own repositories and import the packages. Integrations with other people's protocols (ERC-8004, ANS, HCS, DNS-AID, OASF) live outside too, importing inward, never the reverse.
That is what lets you take one offering without the rest, and what keeps a vertical's vocabulary out of the primitives: the packages are generic trust building blocks; branding, verticals and deployment specifics are the application's job.
Package map and quickstartSee all nine at work in one application.
Game Night uses Identity, Authority, Harness, Edge, Registry Kit, Evidence, Coordination and Operations — in a card room.