Skip to content
AI-assisted purchases, under your rules

Open your store to AI-assisted buying without losing control.

An AI agent is an assistant that searches, compares, or buys by following a person's instructions. You decide what it may do in your store, when it needs approval, and what is recorded if a question arises.

Before you see what is ready · During you apply your rules · After you consult the available evidence.

CONTROL + EVIDENCE
Delegated requestBuy 2 units · EUR 124.00
checkout
01
Agent identitySignature and context evaluated
to verify
02
Policy and riskThe amount requires confirmation
review
03
Merchant stateAuthoritative revalidation in pilot
in pilot
Verifiable exampleTrustReceipt v1.0 · current corpus
JWS · Ed25519

Scope: Trusteed does not process payments, is not merchant of record and does not replace your own records or legal obligations. See full limits

The problem is not allowing agents

Open a new sales channel without losing control of your business

An agent can be authenticated and still act against your commercial interests. Control evaluates the actions it covers and can allow them, ask for review or block them, depending on the surface and the state of the pilot. Where issuance exists, Evidence keeps a signed record of the context and of the decision available at that moment.

01

Who is acting?

A technical identifier does not always explain who operates the agent or what its purpose is.

Verify the context before exposing a sensitive action.How we evaluate context
03

What changed before execution?

Price, stock, seller or terms may no longer match what was approved.

Revalidate authoritative state where the capability is operational.See revalidation in the flow
04

Can you reconstruct what happened?

A dispute requires joining intent, decision, order, delivery and return.

Keep verifiable records and detect evidence gaps.See the sample TrustReceipt
Before, during and after the operation

Before, during, and after: control each important moment.

First you see what is ready. Then you apply your rules while the assistant acts. And, when the capability is available, you can consult a record of what happened.

STATESavailable · in production todayin pilot · with supportplanned · no date
Beforeavailable
TRUSTEED ASSURANCE

Know where to start.

A reproducible diagnosis of what you already have: which control and evidence signals exist today, and what is missing to open the channel.

  • Reproducible scenarios
  • Trust Score with provenance
  • Priority and next step
How Assurance works
Duringin pilot
TRUSTEED CONTROL

Define what an assistant may do.

Verify who is acting, apply your economic limits, ask for human confirmation when it matters and block what you do not accept.

  • Verified agent context
  • Limits, risk and human review
  • Allow, review or block, with a reason
How Control works
Afterin pilot
TRUSTEED EVIDENCE

Consult what was recorded.

When the pilot scope supports it, a signed TrustReceipt helps verify what was recorded and which evidence is still missing.

  • Verifiable TrustReceipt
  • Missing proof, made visible
  • Export aimed at payment providers
How Evidence works
TECHNICAL DIRECTION · NOT AN AVAILABLE FEATUREThe technical substrate connects reliable data sources, decisions and proofs without replacing existing systems.
  1. 01Access and identityobserve
  2. 02Authorisation and statein pilot
  3. 03Policy and decisionobserve
  4. 04Evidence event graphplanned
Proof before promise

A signed receipt lets you verify what was recorded.

TrustReceipt keeps the available context of identity, consent, policies and operation. Its signature lets you check who issued the record and whether it changed afterwards; on its own it does not prove that the underlying fact happened.

It is not a complete lifecycle chain nor a universal legal guarantee. Evidentiary weight depends on the jurisdiction, the process and the concrete evidence available.

See limits and verifier
TRUST RECEIPTexample
receipt_idtr_01JY7E4M8X...A92F
schema
v1.0 · current corpus
subject
checkout.complete
merchant
verified context
decision
human_confirmed
input_hash
sha256: 4b8d...21f0
signature
JWS · Ed25519
live

Individual signed record, issued in production. That every checkout generates one, or any volume, cannot be claimed: the last known receipt is from 2026-08-13 and the last 24h are zero. The lifecycle graph remains planned.

livev1.0Issuance and verification in production; volume not claimable
dormantv1.1Hardening flag disabled
Public Trust Score v4.1

Score for merchant: agenticmcpstore

65.0/ 100 · live reading from the public endpoint

The public v4.1 surface already takes part in ranking and eligibility. It is not a customer review, and it is not a score for Trusteed: it is this merchant's.

CAUTIONconfidence: cold_start
Merchant reliabilityHistory and operationnot_applicable
Agentic readinessSetup, security and policiesmeasured
Agentic evidenceOnly where a real sample existsnot_applicable
METHODOLOGY 4.1.0Confidence with context
  • Bayesian shrinkage
  • Time decay
  • Wilson interval
  • Per-datum quality

The number is computed by v4.1. The per-component breakdown and the methodology document are still v2.0, which is why they do not explain the figure on their own.

How the Trust Score is measured
You do not have to take our word for it

See how the control layer decides.

Change the scenario. The result explains which rule was applied and what has to happen next.

POST /demo-store/mcpsimulation
{  "tool": "complete_checkout",  "amount": 6400,  "agent_status": "verified"}
policy evaluation
DECISIONALLOW WITH CONFIRMATION

The operation may continue. The user confirms before checkout.

rule R030.simple-controlsreceipt queued
Assurance with visible limits

Proof does not replace the merchant's responsibilities.

What we do
  • We connect catalogues, policies and actions.
  • We make operational trust signals visible.
  • We apply limits and confirmations where they are supported.
  • We can issue verifiable records on confirmed surfaces and deployments.
What we do not do
  • We do not process payments.
  • We are not merchant of record.
  • We do not certify that a transaction will succeed.
  • We do not replace the merchant's legal obligations.
Questions before opening the channel

Trust also needs clear limits.

Do I have to change my e-commerce platform?

No. Trusteed is designed as a layer on top of existing systems.

Can an authenticated agent still be a risk?

Yes. Identity answers who is requesting the action; on its own it does not explain whether the sequence is economically abusive.

Does Trusteed record the whole lifecycle?

No. The current corpus contains individual receipts. The causal lifecycle graph is still behind flags that are off: it is not available today.

Does the Trust Receipt guarantee a won dispute?

No. It strengthens traceability and evidence readiness. Evidentiary weight depends on the case, the recipient, the process and the jurisdiction.

Guided beta access

Start with the gap you need to close.

Control and Evidence are offered through guided pilots. Assurance helps evaluate the starting point.

Intelligent Discovery

Find what you need in natural language

Natural-language search over this site, served by NLWeb.

Not available today: the route and search pipeline are deployed, but the embedding step is failing upstream. Try it and you'll see the real error.

For AI agents (machine-readable)

MCP endpoint

POST https://trusteed.xyz/{slug}/mcp

JSON-RPC 2.0 — replace {"{slug}"} with any merchant slug (e.g. demo-store)

Available tools

search_productsget_product_detailsbrowse_categoriescreate_cartpreview_checkoutcomplete_checkout