Open standard · version 1

Agent Terms

One small document in which a business states its terms for AI agents, and every agent reads the same terms before it calls, writes or books.

Read the specification Validate a document JSON Schema

What a business can say

AI agents already phone, write to and book small businesses that have no standard way to set terms. Agent Terms covers the terms a front desk could not set until now:

Whether agents may contact it

Yes, only through the channels it lists, or not at all.

Who is calling, and for whom

The agent says it is an AI, which agent it is, and whom it acts for. Always.

Which channel, when, how often

Channels, phone hours, quiet hours and a limit per agent per day.

Booking rules

Opening hours, what to bring, party size, one booking per customer, no resale.

Deposits and fees

Deposits, card guarantees, no-show fees and cancellation deadlines, told to the customer before booking.

Recording and guardrails

No recording of staff without asking, and guardrails for regulated work: no advice, no diagnosis, the client verified first.

Proof the business is real

A verification tier from a named verifier, and a signature that shows the terms were not altered.

Which agents go straight through

The operator tier the business requires; below it the customer confirms. Nobody is refused for a tier.

How a business publishes its terms

Write the document, by hand or with any tool that produces it. A minimal one is five lines; the examples show a complete one for each sector.
Serve it at /.well-known/agent-terms.json on the business's own site, or point that address to where it is published, or add <link rel="agent-terms" href="…"> to the home page.
Sign it with an Ed25519 key published at /.well-known/jwks.json, so agents can tell it is genuine.

For agent makers

Look for the terms before contacting a business, follow them, and say who you are. The discovery rules, the behaviour an agent owes and the signature check are in the specification. Check your implementation with the test vectors, the validator, or the command line tool, which runs without an account:

curl -O https://snuepi.com/spec/tools/validate-agent-terms.mjs -O https://snuepi.com/spec/tools/spec-validator.mjs
node validate-agent-terms.mjs https://example-business.eu
node validate-agent-terms.mjs --conformance

Neutral and open

Anyone can implement it. Publish, read, validate or build a registry on Agent Terms without permission, registration or payment.
Every agent maker on the same terms. snuepi serves every agent maker the same documents, the same checks and the same tiers. Where treatment differs, it depends only on published criteria every agent maker can meet, never on which agent maker it is. No agent maker gets preferential treatment, and no tier can be bought.
Stable and versioned. Each version has its own address and schema and stays available. A new version is announced here at least 90 days before snuepi stops serving the older one.

Proposing a change

Anyone can propose a change: describe the problem it solves, with an example document, to support@snuepi.com with "Agent Terms" in the subject. Proposals and decisions will be published on this page, and a public issue tracker will be announced here. A change that adds a rule is open for comment for at least 30 days before it goes into a new version.

Versions

Version	Status	Files
Agent Terms v1 (document format 0.2)	Current, 29 October 2026	schema · test vectors · markdown
Agent Policy 0.1 and 0.2	The drafts v1 is based on	schema (compatible with v1)

Licence

The specification text is licensed under CC BY 4.0. The JSON Schema, the examples, the test vectors and the reference validator are licensed under the Apache License 2.0. Implementing the standard needs no licence at all.