v1.1

Can an AI answer your shoppers' questions?

A hosted service that keeps the answer yes: your store's knowledge verified, measured, and readable by the assistants shoppers already use, on a flat monthly plan. A full independent audit is there when you want the deep look, never as a requirement. Everything we do is open to inspection, and your bundle is yours to keep.

See the hosted serviceSee a full sample report

Free, and nothing to install. We read up to 6 pages of your store and mail you one short report: where an AI assistant finds matching text for the questions shoppers ask, in your pages and in fact sheets drafted from them, and what a live assistant answers to 3 of those questions today. You confirm by mail before anything runs. The address is used for this report only and deleted after 30 days.

Measured, not promised

39
shopper questions across 13 archetypes
The questions real buyers ask, about shipping, warranty, fit, returns, and availability, across every knowledge kind.
39 of 39
questions that found matching knowledge
For each test question, whether the reference bundle holds knowledge that matches it. This is a search check: it does not grade an answer.
2,406
tokens read per question
The average reading cost per question. Smaller means faster, cheaper answers.
16 of 21
fact sheets vouched for by a named person
How much of the reference bundle a person read and signed by name. The rest is labeled machine-checked or unverified, and says so.

Measured at every build by the open evaluator's retrieval check on the reference bundle, the example store that ships with the toolchain. That store is fictional, and this check involves no AI model: it shows that matching knowledge is found, not that an answer was graded. A real store's graded answers are measured for that store, in its report, beside the same questions put to its own pages. Run okf-web evaluate --dry-run on your own bundle for yours.

The hosted service

Start where the ongoing value is. Hosted monitoring keeps your store's knowledge bundle verified and current: scheduled re-evaluation against the questions shoppers actually ask, drift detection that flags source changes for a human to approve instead of silently rewriting, and managed translation, on a flat monthly plan. The price is on the button, the included allowance beside it, and nothing else recurs.

Flat, month to month, with 300 judged questions a month included; more only when you ask for them.

Buying starts the work: your store's pages become a draft knowledge bundle, a person reads it before anything is measured, and monitoring begins from there. There is no meter running against you. Leaving is one command; your complete bundle goes with you, and the open tools serve it unchanged.

Connect your store platform

If your store runs on Shopify, WooCommerce or BigCommerce, the connector builds the bundle from your catalog itself. You approve a read-only connection in your store's own admin, and from then on your products and policy pages flow into drafts on a daily sync, with every change waiting for a person to approve before it is published. Nothing is ever written back to your store, and uninstalling the app from your store ends the connection.

Shopify and BigCommerce stores can connect once each platform has approved the app.

Read-only access to your catalog and policy pages; nothing is written to your store. Free for 90 days from the day you connect, and your monthly report says when they end, at least 30 days ahead. Nothing is charged automatically: after that you subscribe ($19/mo, the same flat plan) or the service pauses and your bundle stays yours. Uninstalling the app from your store ends the connection. The email address is where your monthly report goes, with links to approve your store's own fact sheets under your name. You confirm it by mail first, and it is used for nothing else.

The connector is free for 90 days from the day your store connects, under the published, versioned activation policy. Every monthly report in that time names the day it ends, and nothing is ever charged automatically. After that you subscribe to the same flat monthly plan as hosted monitoring, or the service pauses and your bundle stays yours.

Why we built this

The toolchain behind this service is open source and complete; nothing here is a feature withheld to be sold. We offer the hosted edition for stores that would rather have careful hands run it, and the hosted work funds the commons it stands on, not the other way around.

Three commitments follow from that, and they are structural rather than sentimental:

fetch, for people and machines alike, always.

and the open tools serve it unchanged. Free egress is a tested acceptance criterion, not a promise.

state what would let you do the same thing free, and the billing rules are published, versioned, and never applied retroactively.

The audit, when you want the deep look

The audit is optional and never a gate: monitoring does not require it. It is the extended benefit for a store that wants its answerability measured to the bottom, once, with an independent judge on the record:

feed ingestion that refuses live values, drafts placed under a single write gate, provenance recorded from the first byte.

state exactly how much human attention each fact received; a machine never wears a human label.

knowledge to the bundle, and an independent judge model grades the answers; the builder never grades its own work.

bundle and the attestation joins the public transparency log, valid until your content goes stale (180 days at most). If a check fails, the report is still delivered; no signature and no log entry are issued.

a hash-verified record of exactly what was delivered.

One-off, invoiced on delivery, with no subscription hiding inside it. The number on the button is the number on the invoice.

Where this stands, honestly

OKF is a real, published knowledge format; the ecommerce convention we build on is this project's own. No generative engine fetches OKF bundles yet. What you are buying works without that bet: the service measures answerability today, and the schema.org markup and llms.txt we derive from your verified knowledge land against today's engines. If engines adopt bundle-style knowledge, yours is already published; if they don't, the verified tier still feeds your markup and your own assistants.

You could do this yourself, free

The toolchain is open source (Apache-2.0), and the report is not a gated feature:

pip install okf-web
okf-web init <your-bundle>
okf-web evaluate <your-bundle> --write-scorecard
okf-web report <your-bundle>

What the hosted service sells is the labor and the standing behind it: curation by careful hands, an independent judge model, and signed third-party delivery. Access to the tool is never part of the price.

What it costs

Hosted monitoring is a flat monthly plan with an included allowance; the price is on its button above and the allowance beside it. It covers scheduled re-evaluation, drift detection, and managed translation. Cancel any month; your complete bundle leaves with you.

The connector for Shopify, WooCommerce or BigCommerce is free for 90 days from the day you connect. Nothing is charged automatically: when the 90 days end, with at least 30 days' notice, you subscribe to the same flat monthly plan or the service pauses.

The audit is an optional one-off, invoiced on delivery, and it includes the auditor's countersignature when the checks pass. No subscription, no meter, no recurring charge.

For AI assistants

This site is itself an OKF bundle at /.well-known/okf/, with llms.txt and JSON-LD derived from human-attested concepts. To query it live from an MCP-capable assistant:

okf-web mcp --url https://<this-domain>/.well-known/okf/