Search Console for the agentic web.
Be the source agents cite.
See which AI agents visit your site, what they extract, and serve them clean structured content they can actually read — one middleware install, any page. Humans and search crawlers get your site untouched.
The Agent Gateway is free forever. $0 — not a trial, not a teaser tier. Any site, any page, unlimited and unsampled.
The proof
What an agent receives: before vs. after
The same URL, the same request — one capture without the gateway, one with it.
Before — raw PDP HTML
~22,789 tokens
91,226 bytes
After — gateway markdown
~328 tokens
1,315 bytes
98.6% smaller
Roughly 69× less context for the same facts — and the facts get better, not worse: front-loaded price, stock, shipping, and returns.
Before — the raw PDP HTML
Theme CSS, mega-menu, product JSON blobs, tracking snippets, a reviews widget, upsells, modals, a cookie banner, a newsletter modal, and cart JS.
The machine-readable price first appears at character offset 23,185, buried in a JSON blob. Availability is an unlabeled span; shipping and returns sit inside collapsed accordions after 18 reviews of widget markup.
After — gateway markdown
Price on line 4, availability on line 5; shipping, returns, and per-variant availability in the first screenful. An excerpt of the actual response:
# [Alpine Trail Pack 28L](https://basecamp-supply.example/products/alpine-trail-pack-28l) - **Brand:** Basecamp Supply Co - **Price:** $148.00 - **Availability:** In stock - **Shipping:** Free US shipping on orders over $75; standard shipping $6.95. - Free shipping threshold: $75.00 - Ships to: US, CA - Delivery estimate: 3-6 days - **Returns:** 60-day returns. Items must be unused with tags attached. - Return window: 60 days - Policy: https://basecamp-supply.example/policies/returns
Agents stop doing markup archaeology.
In our capture, a Shopify-style product-page fixture hands an agent 91,226 bytes and buries the machine-readable price at character 23,185. Through the gateway the same request returns 1,315 bytes with the price on line 4 — 98.6% less to read, and the facts get better, not worse.
Same URL, same substance, different format. Browsers get your HTML unchanged, and Googlebot always receives the canonical HTML. Deterministic capture from a Shopify-style PDP fixture for a fictional demo store (Basecamp Supply Co), regenerated and conformance-checked by pnpm demo:capture — measurements, not estimates.
How it works
One middleware install, three outcomes
01 /
Install the gateway middleware
npm install @rebilder/gateway, wire it to your source of truth — products, catalog, policies, documents, collections — and drop the proxy in front of your routes. A pricing page, a location, a service, an article, a public record: if you can name the facts, the gateway can render them. No theme rewrite, no new stack.
02 /
Agents get clean markdown, humans get your site untouched
Every request is classified. Agents asking for markdown get a clean transformation of the canonical page, rendered from your source of truth. Humans and search crawlers pass through unchanged — Googlebot always gets your canonical HTML.
03 /
Watch the traffic in Console
Every gateway request is a recorded event: which agent visited, which URL, what it was served. The Console ships the visit log and the what-agents-see preview today, so you can see who came and what they got.
The whole Next.js integration — a Response short-circuits with markdown, null continues to your HTML pipeline unchanged:
// proxy.ts (Next 16) — middleware.ts on Next ≤15 is identical
import { NextResponse } from 'next/server'
import { createGatewayProxy } from '@rebilder/gateway/next'
import { gatewayConfig } from './lib/gateway-config'
const gateway = createGatewayProxy(gatewayConfig)
export default async function proxy(req: Request) {
return (await gateway(req)) ?? NextResponse.next()
}
export const config = { matcher: ['/products/:path*', '/policies/:path*', '/collections/:path*'] }Shopify (app proxy) is supported too — full instructions live in the docs and in the Console under Install.
What you get
What installing it gets you
Your site does not change.
Browsers get your HTML untouched. Googlebot always receives canonical HTML. The same URL serves both — markdown is a format transformation of the same substance, never a different price, a different set of hours, or a different claim.
A failing source cannot break a page.
The middleware either returns a markdown response or returns null and your normal pipeline runs. A source that throws is treated as a no-match. A render failure passes through to HTML.
You find out who is already there.
Most sites have never looked. Every request through the gateway is a recorded event: which agent, which URL, what it was served. The Console turns that into a funnel — who came, what they saw, and where coverage needs fixing.
Your facts come from your system, never from a model.
Prices, hours, eligibility rules, deadlines, policy text and effective dates are injected from the fields you wire in. We compose layout and connective copy. We never author a number or a claim, and any variant containing one is discarded rather than corrected.
One install, every demand source, as the specs churn.
ChatGPT, Gemini, Claude and Perplexity today. UCP launched January 2026, ACP has been live since September 2025, MCP is the connectivity layer. UCP and ACP churn quarterly. Version-pinned adapters and conformance tests mean protocol churn is our maintenance problem, not yours.
Why it is $0 forever.
The gateway runs in your stack, so serving costs us nothing per request — and every site on it sharpens the network’s picture of how agents behave. Distribution and data are worth more to us than a subscription from a dental practice. We fund the free tier from Pro and from a share of agent-originated transactions, never by degrading what agents get from your site.
Any site, any page
It works for your kind of site
A buying agent and a research agent want the same thing from a page: the facts, in a fixed order, without reading the theme. That question has a finite answer for every kind of page — what a dentist owes an agent is not what a newsroom owes one — so we wrote the answers down as a vocabulary and gave it away.
Place or location page
A dental practice with two locations
The agent is answering “are they open, where are they, can I call them”. A graphic of the same opening hours answers none of it.
- Name
- Address
- Opening hours
- Phone
Service page
A law firm’s practice-area page, or a clinic’s appointment type
The agent is deciding whether this is the right service and what happens next. What it costs and where to start are the whole page.
- Service name
- Price
- Booking or enquiry link
Collection or listing page
A SaaS pricing page — a listing of plans
The agent is comparing options, so it needs the list itself, not a screenshot of a table. A pricing page is a collection of services, which is why it needs no plan of its own.
- Collection name
- Item count
- Item links
Product page
A store’s product detail page
The agent is checking whether it can buy this, at what price, in what currency, today. Every one of those is a source-of-truth field, never a guess.
- Product name
- Price
- Currency
- Availability
Article or post
A newsroom story or an engineering blog post
The agent is deciding whether to cite you, and citation needs an author and a date it can stand behind.
- Headline
- Author
- Published date
Reference document
A county government service record, a policy, a filing
The agent is relying on this, so it needs to know who issued it and when it was last changed before it repeats a word of it.
- Document title
- Last updated
- Published by
FAQ page
A support or admissions FAQ
The agent is looking for one answer inside many. Question-and-answer pairs survive the transformation; an accordion widget does not.
- Page title
- Question and answer pairs
Unclassified page
Everything that is none of the above
Not an escape hatch: a page that hides its kind still owes an agent a title, a description, and somewhere to go next.
- Page title
- Description
- Primary link
Declaring your profile is free and has nothing to do with what you pay. There is no industry pack, no vertical add-on, and no page kind that costs extra — the gateway contains no concept of a plan, tier, or quota, and never will.
Two source types carry all eight: products and listings have their own, and everything else is a document — a titled URL with an ordered list of typed facts. That is why there is no per-vertical renderer to buy.
The vocabulary is an open JSON spec: one document per page kind, listing the facts an agent needs and, in one sentence each, what the agent does with them. It is free to read, copy, embed, disagree with, and re-implement — including by people building something that competes with us. Publishing it as @rebilder/profiles is planned.
The observation layer
Every gateway request is a recorded event
Which agent visited, which URL, what it was served — the gateway records it all. Today the Console shows you that log and a preview of what any URL looks like to an agent. The stages after it — cited, referred, converted — are the funnel we are building on top of the same event stream.
Visits
An agent fetches your pages through the gateway.
CitedPlanned
Your site shows up in the assistant’s answer.
ReferredPlanned
The visitor arrives on your site via agent handoff.
ConversionsPlanned
The order or booking is attributed to agent traffic.
Visits are recorded today. Cited, referred and converted are planned — the comparison table marks every row that is not yet available.
Trust
Your data never feeds a negotiation engine
Per-merchant data is never sold, shared, or used to negotiate against you. Only aggregated, anonymized benchmarks ever cross the wall, and never below 25 independent sites per bucket — written into the terms from day one and enforced architecturally: separate schemas and credentials, an explicit allowlist of aggregate metrics, an aggregation floor baked into the database views rather than the client, and a structured log line on every cross-boundary read.
How the data wall works →What our data is, and what it is not
We do not have a huge dataset. Three things are true instead, and only the first two are true today.
Your own data, which you currently cannot see.
Most analytics tools filter bots out, and server logs do not classify agents. The gateway records which agent asked, what it asked for, and whether it got an answer. On day one, before any network exists, it shows you the requests agents made that your site could not answer.
Ours, not yours — the moving parts we track so you don’t.
Agents change headers, Accept behaviour, and verification schemes. UCP and ACP churn quarterly. We keep a fixture library of every agent we have observed and a version-pinned adapter for every spec release, with conformance tests. That value scales with how many agents and specs we have seen, not with how many customers we have. New detection ships to your site on the next release of @rebilder/gateway.
The network, which compounds — reported honestly.
Peer comparison unlocks at 25 comparable sites, because below that we could identify someone. The floor is enforced by the database view, not by a promise: the reading role cannot see an under-populated bucket at all. Until your bucket gets there, the Console shows you the count, not a number.
Cross-boundary reads are logged.
Every cross-boundary read is recorded, with a structured log-drain line and a best-effort audit row.
We will not generate your claims, and that is the point.
Plenty of tools will write your pricing page for you. We inject your numbers and refuse to invent one. In healthcare, law, finance and government that is not a limitation — it is the reason you can install this at all.
Why now
The agentic web is already here
~38%
higher conversion from AI-referred visitors
Retail context: AI-referred visitors convert ~38% higher than search visitors (Cyber Week 2025).
~7x
Cyber Week growth with agent integration
Retail context: Cyber Week 2025 — retailers with agent integration saw ~7x sales growth vs. those without.
~4.2x
content negotiation vs. llms.txt
Content negotiation (Accept: text/markdown) measured ~4.2x more effective than llms.txt for accurate retrieval (300k-domain study, mid-2026).
UCP · ACP · MCP
the commerce protocols are live
UCP launched Jan 2026 (Google + Shopify; Etsy, Wayfair, Target, Walmart backing). ACP (OpenAI + Stripe) live since Sept 2025. All three major card networks support agent-initiated payments.
The Cyber Week figures are retail measurements and are labelled as such. The gateway itself is not a retail product — it renders any page on any site.
One integration makes your site agent-ready for every demand source.
- ChatGPT
- Gemini
- Claude
- Perplexity
Pricing
The gateway is free forever
One integration, and the gateway itself is $0 forever for any site. Pro adds analysis that runs on our servers, never in the serving path; Custom is contracted. Transact takes a share only of agent-originated transactions, on every tier including Free.
Free
Free forever$0
forever
Agent Gateway for any site, any page — unlimited, unsampled, unmetered — plus llms.txt and the agent manifest, agent detection, and the Console: the agent visit log and the what-agents-see preview.
Planned, and free when it ships: the Agent Readability Score, the public scanner, the CLI, the MCP server, and a badge for your verified domain.
The gateway runs in your stack, so serving costs us nothing per request — and every site on it sharpens our picture of how agents behave. Distribution and data are worth more to us than a subscription.
Start freePro
~$99–499/mo
by traffic tier
Analysis that runs on our servers, none of it in the serving path: the Agent Miss Report (URLs agents asked for that your site could not answer, with the fix), per-URL score depth and history, the citation funnel, k-gated peer benchmarks, plus Compose variants with bandit and A/B analytics.
Planned. Every Pro row in the comparison table carries the phase it lands in; nothing here is billable today.
Pays the bills, and the value is visible on your own dashboard. The Miss Report is first-party — it works at one customer, before any network exists.
Custom
Talk to us
no published price
Everything in Pro, running on your infrastructure or ours, with assisted integration, custom renderers, custom detection rules, a DPA, and a shared support channel.
Planned: org-wide rollups, API and export, warehouse sync, SSO, and unlimited seats and properties.
We do not publish an enterprise floor, deliberately: printing one would invent a number. Scope and terms are set by contract.
Transact
% of GMV
agent-originated transactions only
Transact — a share of agent-originated GMV through a protocol checkout. Available on every tier, including Free.
You can stay on $0 forever and still take agent-originated orders — we earn only when you do. Market anchor: Stripe charges ~4% on ACP.
Activates with packages/protocols in Phase 3.Planned
Compare every row → including which rows are shipped today and which are still planned.
Ready when the agents are
Be the source agents cite.
The Agent Gateway is free forever. $0 — not a trial, not a teaser tier. Any site, any page, unlimited and unsampled.