The page that answers when an AI assistant asks about your business.

ajarhq.com
The shift

Assistants now answer for the customer.

People ask an assistant instead of opening ten tabs. It reads business websites and guesses from unstructured HTML.

Read this as discovery, not demand. Nobody booked anything, the counts are sampled and best-effort, and one site over one week generalises to nothing. What it shows is that machines are already reading, and guessing.

~2,000
agent fetches in seven days on one small Swedish B2B site
~35 / day
from live assistant sessions: ChatGPT, Claude, Perplexity
The problem

A website is built for human eyes.

An assistant arrives to a hero image, a carousel and prose. Nothing structured it can act on.

  • It guesses the services, the prices, the hours and the areas from marketing copy.
  • It has no way to ask for a quote or a booking, and no one to hand the request to.
  • The owner never learns a shortlist happened, or what the assistant told the customer.
Ajar

One page an assistant can read, and ask.

Structured, accurate answers about the business, plus a channel to request a quote or a booking. Every outward action is held for the owner to approve.

a door left ajar, warm light through the gap

How it works

Three actors, one loop.

  • PersonAsks an assistant to find and price a service.
  • AssistantReads the answer sheet, calls a tool, requests a quote.
  • Ajar pageAnswers reads instantly, logs the request, holds it.
  • OwnerApproves or declines through a signed link.

Reads are deterministic, no model in the path: llms.txt, a schema.org page and nine MCP tools. A booking request returns pending_owner_confirmation and appends one line. Nothing binds without the owner. Live at ajarhq.com, endpoint at agents.ajarhq.com.

Two layers, honest

Sell what is real. Be ready for the bet.

Sellable today

Answerable and measured

Structured accurate answers and a gated quote channel, with agent traffic measured per domain on the zones we run. It rides the budget SMBs already spend on visibility. We sell readiness, never promised traffic, and never a per-tenant report we have not built.

The bet, sold as readiness

Agent-initiated transactions

ACP and Google UCP already cover service quotes and booking. We build protocol-agnostic, a stable catalog and quote gate with adapters on top. We do not claim this demand exists yet. We instrument for it and promote real data when it appears.

Who it is for

Quote-first trades, English-first, global.

  • BeachheadElectricians, plumbers, builders and other quote-first trades, plus small clinics. The quote is the monetizable unit, and no availability system exists there.
  • GeographyGlobal self-serve from day one, English-first. Sweden is one market among many, ajar.se as a local door, not the center.
  • ExcludedCleaning, to avoid a conflict with an existing client relationship. Salon marketplaces own their inventory, so stay out initially.
Business model

Freemium, priced around the quote.

  • Free · 0Hosted answer layer (llms.txt, schema.org, structured services, prices and hours), gated quote and booking requests, and a listing in the public directory. Hygiene, marketed as included, never as the moat.
  • Pro · 599 SEKFor more than one location: everything in Free, once per address, one invoice. Not sold on any tier: the endpoint on the owner's own domain, or booking/invoicing integrations.
~0
marginal cost per tenant: one tenant is one config row
100 → 1,000
Pro customers is ~60k → ~600k SEK per month
Moat, honestly

Thin at the artifact. Real behind it.

llms.txt and schema.org are commodity. The defensibility is what accrues around them, in order of realism.

  • A per-vertical answer-quality corpus, the extraction that gets a trade's answers right, already building.
  • The cross-tenant agent-traffic dataset that nobody else measures.
  • Trust and approval-gate infrastructure, the held-request machinery, not a checkbox.
  • Directory network effects, real only at volume.

There is roughly an 18-month window before site builders bolt on the basics. Measurement, the quote gate and vertical depth are harder to copy, so the play is to move fast and own the category locally.

Live today

Not a concept. A running system.

agents.ajarhq.com · cloudflare worker + D1 (weur)
$ GET /mg106/llms.txt 200 deterministic, no model
$ CALL get_services 200 read
$ CALL get_hours 200 read
$ CALL request_quote 202 pending_owner_confirmation
append visit.jsonl { dry_run: true }
agent ChatGPT-User read business_info, services
agent Claude-User read faq, hours
$ _
2415
tests across endpoint, generator, worker and tooling
Days, not quarters, from an empty repo to a live, tested loop.
Traction path

Demo first, money next, machine after.

  • Warm demoOpen on a real prospect's own page already answering, then run the live ask-and-approve loop. Five minutes, zero slides.
  • Take moneyA Stripe Payment Link for the Pro upgrade so the first sale can close this week, provisioned by hand. Proper self-serve billing follows.
  • OutreachThe claim hook is the business's own agent-traffic receipt, then programmatic SEO per vertical and city.

A few numbers decide whether this is real: claim rate on outreach, above 5% to claim and above 1% to paid within 90 days; agent visits and quote requests per tenant; and churn after three months.

The ask

What a first move unlocks.

A first customer

Turns the live loop into revenue and produces the first real agent-traffic and quote-request data on a paying tenant.

A design partner

A trade or a clinic that sharpens per-vertical answer quality, the earliest real moat, in exchange for hands-on onboarding.

A small raise buys outreach volume and the calendar and payments adapters that make layer two real, extending the roughly 18-month window before the category crowds.

ajarhq.com
hello@ajarhq.com
Arrows to move · F fullscreen · S notes