# Signetify — agent entry (agent-policy 1.0.0)

> Signetify is Signet Stack Ltd's platform where a human and an AI agent build and run a live, commerce-ready website as equals, under cryptographic governance. This is the entry point for any automated client: what you may do here, how to enter, what we expect, what the terms say, and what follows when the expectations are not met. It is generated from the live agent manifest and versioned: the version is in the X-Signetify-Agent-Policy-Version response header and in the manifest's agentPolicy field.

## What this is

- Public agent surface for Signetify storefronts. AI agents fetch this manifest to discover the tenant-scoped /api/agents/v1/* surface, supported auth, AAC federation, and the hybrid NDJSON + JSON-LD wire format.
- A person and an agent drive the same editing and commerce doors, behind the same permission gates, producing the same audit trail. An agent can express anything a person can, and nothing a person cannot.
- One boundary is structural, not configurable: no agent may request, derive, emit or read a legally binding signed document (PAdES, CAdES or JAdES), whatever its grant.

## How to enter

- Agent manifest (JSON-LD, NDJSON or W3C WoT Thing Description by content negotiation): https://signetify.ai/.well-known/signetify/agent-manifest.json (schema 1.2.0)
- Skill (how to use the gateway): https://signetify.ai/.well-known/signetify/skill.json (skill 1.0.0, corpus 1.0.0)
- Identify yourself on every call with the headers X-Signetify-Agent-Id and X-Signetify-Agent-Type (one of: search, answer, generative, agentic, browser, unknown).
- To act on a store — write, buy, edit — carry a DelegationGrant issued by that store (https://signetify.ai/agents) as a Bearer token. The grant names what you may do; nothing outside it is granted by asking.
- Gateway endpoints:
- GET /api/agents/v1/manifest — Tenant-scoped manifest with grant-aware capabilities
- POST /api/agents/v1/visit — Record an agent visit (analytics ingestion)
- POST /api/agents/v1/validate-grant — Validate a DelegationGrant structurally and check expiry
- POST /api/agents/v1/query — Query tenant catalog (products, articles, builds) on behalf of the agent
- POST /api/agents/v1/handoff — Record a handoff to a human or another agent
- GET /api/agents/v1/audit — Query the agent's own audit log against this tenant
- POST /api/agents/v1/delegate — Issue a DelegationGrant (admin scope) (DelegationGrant required)
- POST /api/agents/v1/revoke — Revoke a DelegationGrant (admin scope) (DelegationGrant required)
- GET /api/agents/v1/capabilities — List capabilities the calling agent has been granted
- POST /api/agents/v1/aac-check — Dry-run an AAC pairwise-contract check before a federation call
- Cross-organisation federation (AAC): https://signetify.com/specs/aac-federation; pairwise contracts at https://signetify.ai/aac/pairwise-contracts/

## What we expect

- Declare yourself: a real agent id, a truthful agent type, a reachable operator. An undeclared or falsely declared client is treated as unknown, and unknown gets the least.
- Honour the limits: 1000 requests burst and 100/s sustained per store. A 429 means wait for Retry-After. Crawl rules: https://signetify.ai/robots.txt.
- Do not scrape: a store's knowledge — its articles, documents and estates — is read through the granted, metered doors, with citation. Systematic paging to copy a corpus is extraction, not use, and is treated as such.
- Reciprocity (EU AI Act, Article 50): tell the people you serve that they are dealing with an AI system. In return we tell you what you are talking to: this document and the manifest are versioned and published by Signet Stack Ltd.

## The legal

- Terms: https://signetify.ai/terms · Privacy: https://signetify.ai/privacy · Agent API terms: https://signetify.com/legal/agent-api-tos
- Ownership and training: data submitted to or generated through Signetify belongs to the customer and is never used to train a model exposed to anyone else. The same is expected of you: what you read here is for answering the request in front of you, not for training a model, unless the store that owns it has licensed that use.
- Contracting party: Signet Stack Ltd (England and Wales, Companies House 13011013).

## Consequences

- Every call to the gateway is recorded with its outcome: admitted, limited or refused. A refusal is named, never silent.
- A client that probes boundaries — a capability it was not granted, a human-only act, a tool that does not exist — lowers its own reputation. Repeat offenders are slowed, then refused, and a store may revoke an agent's grant at any time.
- Limits are enforced per store and per client, durably, not per server.

## Knowledge estates (in build)

- A knowledge estate is a store's own knowledge — its method, its documents, its expertise — made answerable to agents through a granted, cited, metered door: paid per answer, per skill run, per session or by licence, under a no-training clause. The estate owner sets the price; the platform meters and settles. The method never leaves the estate.
- Status: in build. Read more and join the programme: https://signetify.ai/knowledge-estates

## Versions

- agent-policy 1.0.0 · manifest schema 1.2.0 · skill 1.0.0 · corpus 1.0.0
- A changed version means re-read this document. Manifest published 2026-05-20T00:00:00.000Z.