Definition
What is
Agentic Primitives?
Agentic Primitives is an open-source trust substrate for AI agents. It answers, for every act an agent takes on someone’s behalf, who is acting, whether they may do this, and what they did — with artifacts a counterparty can verify without trusting the platform that ran the agent.
01The three primitives
What every act can answer.
- 01
Who is acting?
Identity that survives the runtime
Every person, organization and service in your application is a Smart Agent — an on-chain account that can verify signatures, hold value and execute logic. Names, cards, registry entries and DID documents are projections of it, never the identity itself. Passkeys and keys rotate; the agent does not.
- 02
May they do this?
Authority that is a grant, not a token
Permission to act is a delegation the principal signs, narrowed by caveats — a payee, a ceiling, one intent, a time window — and revocable in one transaction. It is verified before every step, again after every approval, and again on chain when value moves. No cached verdict ever authorizes an act after the authority behind it is gone.
- 03
What did they do?
Evidence the owner carries
Every protected step leaves a receipt — which grant, which decision, which transaction, which playbook — into a hash-chained log and a W3C PROV-O graph held in the owner's vault, not in a vendor's trace store. A receipt travels because the owner carries it, and a counterparty can check it without the runtime's cooperation.
Intelligence may be probabilistic. Authority must not be.
02Disambiguation
What Agentic Primitives is not.
LangGraph, the Microsoft Agent Framework, Dapr Agents and their peers orchestrate a run. Agentic Primitives is the layer they can stand on: identity that survives the runtime, authority the runtime cannot forge, evidence the runtime does not own.
Many registries will exist. In most designs the registry signs the agent’s record; here the agent signs its own record and a registry can only say it listed it. The kit is what registries are built from.
OpenAI uses the phrase for Skills, a hosted shell and server-side compaction — features of one vendor’s agent API that make a long-running agent work better. Those are building blocks of a runtime. This project is the substrate under any runtime: the identity, the grant and the receipt belong to the person, not to the API that hosted the run.
Intelligence may be probabilistic; authority must not be. The planner proposes. It is never consulted about whether a step is allowed.
03In one table
The substrate, in a sentence each.
| Primitive | What it is | Standard | Where |
|---|---|---|---|
| Smart Agent | A person, organization or service as an account that can sign. Names, cards and registry entries are projections of it. | ERC-4337 · ERC-1271 | Day 2 |
| Delegation | Permission as a signed grant with caveats, revocable in one transaction — not a token that is checked once. | ERC-7710 | Day 6 |
| Mandate | A delegation bound to one intent — payee, amount, resource, outcome. The confirmation is the signature. | Digest-binding + payment caveats | Day 8 |
| Harness | Planner proposes, mandate authorizes, executor acts, receipt proves. Verified per step, outside the model. | A2A · MCP | Day 16 |
| Receipt | Evidence of every protected act, in the owner’s vault, recomputable by a counterparty. | W3C PROV-O | Day 19 |
| Ontology | One description of the domain that every layer binds to by IRI. Meaning flows through it; authority never does. | RDF · SHACL · JSON-LD | Ontology |
Read the argument, then run it.
Twenty-one days, one idea each, and six live apps that exercise every layer.