# Agentic Primitives > Rails, not throttles. Authority for agents that the agent cannot exceed. Identity, authority and evidence designed as one system — not stitched from ten products. Positioning: the industry answers agent risk with throttles — containment (sandboxes, restricted modes, kill switches), supervision (permission prompts, a second model grading the first) and platform governance (one vendor's ACLs, reversibility and lineage). Each bounds something; none bounds authority, so none can say under whose authority an act happened. Agentic Primitives bounds authority: a grant the principal signs, caveats that are enforcer code, verification on every step outside the model, revocation in one transaction, a receipt the owner carries. - Canonical site: https://agenticprimitives.dev - Public source (the substrate, Ring 0): https://github.com/agentictrustlabs/agentic-primitives - The estate product (Home, agent runtime, vault, edge, Home MCP; deploys Faithnet): https://github.com/agentictrustlabs/ap-home - The town (registry, public graph, naming, chain access, the town portal — the services every estate on one chain shares): https://github.com/agentictrustlabs/ap-town - Four scales, one repository each: substrate (agentic-primitives) → estate (ap-home) → town (ap-town: estates on one chain and the services they share — naming, public graph, registries) → federation (ap-federation: towns across chains on public ground). Each depends only on the one below; authority never leaves the chain it was signed into. - npm: https://www.npmjs.com/org/agenticprimitives - License: MIT - Author: Richard Pedersen (Agentic Trust Labs) — https://www.richcanvas3.com/about-us - Domain ontologies (Rich Canvas): https://www.richcanvas3.com — services https://www.richcanvas3.com/our-services, CommerceCore https://www.richcanvas3.com/commercecore-ontology, did:aa https://www.richcanvas3.com/didaa, writing https://www.richcanvas3.com/muses-of-an-architect - Reference Home (sign in, ceremonies, vault): https://faithnet.me - Skills registry (playbooks by digest, ontology graphs): https://skills.faithnet.io The one-sentence claim: when an AI agent spends money or touches data on someone's behalf, this stack can prove who allowed it, what was allowed, and lets them take that permission back — without trusting any single vendor. Intelligence may be probabilistic. Authority must not be. ## Product pages - [What is Agentic Primitives?](https://agenticprimitives.dev/what-is-agentic-primitives): the definition, the three primitives, and how it differs from OpenAI's "agentic primitives" (Skills, Shell, Compaction) and from agent frameworks and registries - [Thesis](https://agenticprimitives.dev/thesis): build an ecosystem where every participant expands what everyone can accomplish — the competing bet is to coordinate whole journeys through independent participants while each retains control of its contributions; healthcare, fitness, faith, finance and travel; the ecosystem builder as customer; platform vs network-of-integrations vs shared domain foundation; the strategy (shared foundation, domain layer, ecosystem experience, delivery sequence); thirteen bets each with its test; eleven design principles; the layered ontology; initial proofs and measures per domain; where the bet can fail - [A critical review](https://agenticprimitives.dev/thesis/review): the project's own critical reading of the thesis on its merits — a new offering, conceded; the three testable claims inside the vision; the platform as the competitor and estates as the difference; the five markets on their premises; the thirteen bets on their reasoning; the eleven principles as principles (consistent, thin on governance); what the thesis does not say; what would prove it; what a doubter is right about - [About](https://agenticprimitives.dev/about): Agentic Trust Labs, the founder, boilerplate for press - [Home](https://agenticprimitives.dev/): positioning, stitched vs seamless, nine offerings - [Platform](https://agenticprimitives.dev/platform): Identity, Authority, Harness, Edge, Registry Kit, Evidence, Coordination, Ontology, Operations - [Assessments](https://agenticprimitives.dev/audits): published production-readiness assessments (self-assessments, labelled). Latest: [Production Readiness Assessment 2026-09-18](https://agenticprimitives.dev/audits/production-readiness-2026-09-18) — PDF https://agenticprimitives.dev/audits/2026-09-18-production-readiness-assessment.pdf - [Demos](https://agenticprimitives.dev/demos): every live app on the estate — the Home (faithnet.me), your agent inside Claude (Home MCP), Game Night, Gather27 (find / host / ops), Field, the skills registry — with sign-in personas, a script and what each proves - [Architecture](https://agenticprimitives.dev/substrate): Smart Agents, mandates, harness sequence, two knowledge tiers - [Versus](https://agenticprimitives.dev/compare): the stack you would buy — 29 products in seven bands (Auth0, Privy, WorkOS, Safe, Pimlico, Turnkey, MetaMask DTK, Cerbos, OpenFGA, HashiCorp Vault, Postgres+RLS, Veramo, EAS, DocuSign, Box, LangGraph, Temporal, MCP SDK, a2a-js, XMTP, Stripe, Kong, ENS, an ERC-8004 registry, Tailscale, Datadog, TopBraid …), the seam each band introduces, and the 83 packages + 38 contracts per chain that cover every row on one identity - [Versus · the composition](https://agenticprimitives.dev/compare/composition): eight trust-substrate concerns against six peer families — agent frameworks (MAF, ADK, LangGraph, Dapr, Agno, Strands, Mastra, Pydantic, CrewAI, OpenAI SDK, Buzz), Web2 IAM (Auth0, Okta, Entra), smart accounts + delegation (MetaMask DTK, Smart Sessions, Lit Vincent, Safe, Zodiac, Coinbase, Kite), the ERC agent stack (8004, 8001, 8183, 8196, 8273, 8226), registries (ERC-8004, AGNTCY, NANDA, ANS), data/provenance (Inrupt, PROV-AGENT, OTel). Nobody has the composition. - [Versus · throttles](https://agenticprimitives.dev/compare/throttles): containment vs supervision vs platform governance vs authority-as-a-grant - [Versus · who could assemble it](https://agenticprimitives.dev/compare/assemble): competitive analysis of the horizontal agentic market, dated September 2026 with a projection to early 2027 — the capability map — nine areas, some forty functions, and the package that covers each (identity, authority, harness and orchestration, memory and context, triggers, edge and protocols, registry, evidence and credentials, coordination, messaging and collaboration, library and content, Home, ontology, vault and private storage, public knowledge, KMS, developer kit); the field dated (Okta Agent SSO / XAA GA Aug 2026, Entra Agent ID and Agent 365, AWS AgentCore Policy and Consent Portal, Cloudflare OS / Wallets / cloudflare.pay / Identity-Aware AI Gateway, Google AP2, Anthropic Managed Agents, OpenAI Agents SDK, Visa TAP and Mastercard Agent Pay, ERC-8004 and x402, Keycard / Scalekit, AGNTCY / NANDA / MCP registry, LangSmith Deployment); four assemblies that could produce something similar (Cloudflare-native, enterprise identity, hyperscaler platform, Web3) with what each would have and lack; a function-by-function scorecard across all areas; sixteen deficiencies dated to the window; what no assembly produces - [Versus · frameworks](https://agenticprimitives.dev/compare/frameworks): per-framework take/differ and the scorecard - [Versus · Web3](https://agenticprimitives.dev/compare/web3): the Web3 trust-substrate landscape and the ERC stack - [Versus · where we lose](https://agenticprimitives.dev/compare/honest): where peers are ahead today, what we refuse to take - [Game Night](https://agenticprimitives.dev/examples/game-night): complete third-party example (gamenight.faithnet.io) - [Bible translation](https://agenticprimitives.dev/examples/bible-translation): translating the Bible from the source with an ontology at the centre — the loop on John 3; the T/R/C/A-box and ten layers; how the ontology generates the skills, is the knowledge base (referents, divine persons, the explorer snapshot), declares the plan and keys the harness; the licence per layer; a detailed comparison with committees, the SIL stack, Greek Room/Clear, unfoldingWord and the Anselm Project Bible - [Radar design](https://agenticprimitives.dev/writing/designing-a-radar-with-an-ontology): designing a tracking radar from open physics — the tracking and signal ontologies, the signal skill artifacts (library, ambiguity, receive chain, selection policy) and the characteristics they encode, a planted-error harness with development-set scores, and a workbench; synthetic targets; no classified parameters; no engagement advice - [Ontology](https://agenticprimitives.dev/ontology): the Agentic Trust upper ontology and the Texas hold'em domain ontology; how every layer binds to one T-box by IRI; meaning flows, authority never does. Notes: [Membership, role, and stewardship](https://agenticprimitives.dev/ontology/membership-role-stewardship) - [Architecture notes](https://agenticprimitives.dev/architecture): how the substrate is deployed at four scales — the substrate packages; an estate (Home, agent runtime, vault, edge, Home MCP, RPC gateway, estate chain, KMS tenant) and who lives there; the town the estates on one chain share (agent naming, the public graph, registries, the KMS pilot); the federation across chains (public ground, one principal across chains from ERC-8004 and ENS v2, what crosses a chain and as what, what ten chains would break); one act across an estate boundary; Project NANDA's twelve layers against the estate. Notes: [Federation and anonymous credentials](https://agenticprimitives.dev/architecture/federation-and-anonymous-credentials); [Operating Agentic Primitives — the ladder](https://agenticprimitives.dev/architecture/operating); [Substrate, estate, town, federation](https://agenticprimitives.dev/architecture/scales); [The estate and the federation](https://agenticprimitives.dev/architecture/estate); [Chains and estates](https://agenticprimitives.dev/architecture/chains) - [Developers](https://agenticprimitives.dev/developers): package map, standards, status - [Build](https://agenticprimitives.dev/build): say what application you want; point a coding agent at github.com/agentictrustlabs/agentic-primitives + /pokernight and the faithnet estate; test as six demo people (alice bob carol dave elena nathan); the flow; the gates - [Ontology namespaces](https://agenticprimitives.dev/ns): IRI prefixes (`https://agenticprimitives.dev/ns/#Term`) ## Offerings - Identity: One canonical agent per person, organization and service. - Authority: Scoped, revocable grants between agents — the core primitive. - Harness: Ask → Intent → Mandate → Plan → Verify → Receipt. - Edge: Admission at the boundary — HTTPS required, mTLS optional, authority always. - Registry Kit: Build your own registry; be what registries are built from. - Evidence: Receipts, provenance and audit that outlive the platform. - Coordination: Work between agents: endeavors, plans, commitments, decisions. - Ontology: How the domain is shaped — bound by IRI, checked at build. - Operations: Run it: KMS delegates, deploy recipes, live gates, conformance. ## Writing (full text on this site; LinkedIn is optional) - [Two machines translated the same Greek into the same language. Deciding which did it better turned out to be the hard part.](https://agenticprimitives.dev/writing/what-counts-as-a-better-translation) - [Auditing the American English translation: ontology, agent harness, tests, and the workbench](https://agenticprimitives.dev/writing/auditing-the-english-bible-translation) - [Tar Melbene has no Bible, and the model does not know the language.](https://agenticprimitives.dev/writing/naba-a-language-the-model-does-not-know) - [A machine designed a tracking radar. The hard part was making sure it didn’t invent the physics.](https://agenticprimitives.dev/writing/designing-a-radar-with-an-ontology) - [A machine translated the Bible from Hebrew and Greek. The hard part was making sure it didn’t cheat.](https://agenticprimitives.dev/writing/translating-the-bible-with-an-ontology) - [Your personal agent is the new walled garden](https://agenticprimitives.dev/writing/tear-down-the-walled-garden-for-agents) - [Rails, Not Throttles: The AI Model We Need to Change Isn’t the LLM](https://agenticprimitives.dev/writing/rails-not-throttles) - [One Description of the World: How an Ontology Becomes a Rail](https://agenticprimitives.dev/writing/one-description-of-the-world) - [The Ask at a person’s agent: from the words to the receipt](https://agenticprimitives.dev/writing/agent-ask-harness) - [HeartCoach Ask: an app-side harness, step by step](https://agenticprimitives.dev/writing/heartcoach-ask-harness) - [Proposal: run the coach through the harness](https://agenticprimitives.dev/writing/coach-through-the-harness) ### The missing layer (21-part series) - Day 1: [The agentic web has a missing layer — and it isn’t discovery](https://agenticprimitives.dev/writing/the-missing-layer) - Day 2: [Your agent’s identity should be able to sign](https://agenticprimitives.dev/writing/identity-that-signs) - Day 3: [Names, cards, registry entries: all projections](https://agenticprimitives.dev/writing/projections) - Day 4: [Credentials rotate. Identity doesn’t.](https://agenticprimitives.dev/writing/credentials-rotate) - Day 5: [Every agent is a person, an organization, or a service](https://agenticprimitives.dev/writing/person-org-service) - Day 6: [Delegation is the artefact, not the token](https://agenticprimitives.dev/writing/delegation-is-the-artefact) - Day 7: [A grant that is checked once is a grant that is cached](https://agenticprimitives.dev/writing/checked-once-is-cached) - Day 8: [The mandate: binding authority to one intent](https://agenticprimitives.dev/writing/the-mandate) - Day 9: [“Send Alice 10 dollars” — confirmation is a signature](https://agenticprimitives.dev/writing/confirmation-is-a-signature) - Day 10: [The service that acts as an agent must never be that agent](https://agenticprimitives.dev/writing/the-key-is-a-delegate) - Day 11: [Trust is a graph, not a score](https://agenticprimitives.dev/writing/trust-is-a-graph) - Day 12: [Be what registries are built from](https://agenticprimitives.dev/writing/what-registries-are-built-from) - Day 13: [Resolution is not authority](https://agenticprimitives.dev/writing/resolution-is-not-authority) - Day 14: [Two tiers of knowledge that never meet](https://agenticprimitives.dev/writing/two-tiers) - Day 15: [The vault is the record; everything else is a cache](https://agenticprimitives.dev/writing/vault-is-the-record) - Day 16: [Planner proposes, mandate authorizes, executor acts, receipt proves](https://agenticprimitives.dev/writing/planner-mandate-executor-receipt) - Day 17: [Coordination is not orchestration](https://agenticprimitives.dev/writing/coordination-vs-orchestration) - Day 18: [Don’t build your app twice.](https://agenticprimitives.dev/writing/dont-build-your-app-twice) - Day 19: [Receipts that travel with the agent](https://agenticprimitives.dev/writing/receipts-that-travel) - Day 20: [HTTPS required, mTLS optional, admission always](https://agenticprimitives.dev/writing/https-mtls-admission) - Day 21: [Rails, not throttles](https://agenticprimitives.dev/writing/rails-not-throttles-day-21) ## Optional - RSS: https://agenticprimitives.dev/rss.xml - Sitemap: https://agenticprimitives.dev/sitemap.xml - Full writing index: https://agenticprimitives.dev/writing ## Do not cite as - A token or a public-chain product (contracts deploy to any EVM; no token to buy) - An LLM framework (the planner is a port) - A registry, a wallet, or a Home (those are products built FROM these primitives)