No training required

Messaging that never goes off shift.

You have already answered the same forty questions a thousand times. Linkbase reads those conversations, learns your answers and your voice from them, and then covers every channel at every hour. There is no FAQ to write and no tone-of-voice wizard. Your first screen is a review queue, not a blank form.

Try the AI Book a walkthrough

Linkbase · Unified inbox

AirbnbVRBOBookingEmailSMS

Six places to watch

Wifi password

Parking

Check-in time

Directions

+ 35 more, handled

Lock offline · 5B

Airbnb2:14 AM

Maya Chen

Harbourview Loft · Fri–Mon · 2 nights left

Could we check out at 1pm on Sunday instead of 11?

Read before replying

This reservationProperty guideHouse rulesWifi + accessUpsell menuThread history

Of course, 1 hour is $25, or 1:00 PM for $40. Want me to add it?

Paid $40 · your Stripe

Door code valid to 1:00 PM

39 handled · 1 escalated to you by SMS

You have already trained it

Most AI tools hand you a blank knowledge base and call it customization. Linkbase hands you a filled one and asks you to check our work. Scroll to follow the whole loop: what we read on day one, what we pull out of it, what you approve, and why it keeps getting better after you stop looking.

Step 1 of 6 · Connect

It reads the conversations you have already had

Connect your property system and Linkbase does not just pull bookings forward. It reaches backward through your past reservations and pulls the full message thread from each one.

What's on screen

Your PMS connecting, then the backfill walking back through past stays
Every past reservation's message thread pulled and stored
Reservations we already have are skipped without an API call

Under the hood

Normal sync only covers bookings from today forward, so a new host would start with zero past conversations, which is exactly the data the learning needs. The backfill closes that gap deliberately slowly, because it shares an API with everyone else.

Step 2 of 6 · Mine

Your own answers, broken into facts that survive retrieval

You have answered the same forty questions a thousand times. Linkbase turns those answers into atomic facts: one fact about one thing, every pronoun resolved, so it still makes sense months later with no conversation attached.

What's on screen

A real thread you already answered
One sentence becoming two facts, because two questions will ask for them
Absences called out, with the source stamped on every fact

Under the hood

We hunt negative facts on purpose. No AC, no elevator, no early check-in. Models under-extract these because they read as no information, and they are the most valuable facts in the system: they are what stops an AI promising a guest something that does not exist.

Step 3 of 6 · Approve

Your first screen is a review queue, not a blank form

Onboarding ends on a list that says here are twenty things we learned about your business, approve the ones that are right. Each one shows the actual guest conversation it came from. Your job is checking our work, not writing documentation.

What's on screen

Extracted knowledge waiting for you, grouped and ranked
The verbatim quote and the thread it came from, as evidence
Approving publishes it. Nothing reaches the live knowledge base on its own

Under the hood

Facts land in an append-only ledger with a stable identity, so re-reading a conversation updates the same row instead of duplicating it, and a second conversation saying the same thing raises its corroboration count. A fact you stated yourself is never dropped by a ranking.

Step 4 of 6 · Your voice

One file that holds how your business sounds

Every business gets a persona file: tone, phrasing to prefer or avoid, how you refer to yourself, the sign-off you actually use, and standing rules like never offer discounts. We write the first draft from up to sixty of your own messages.

What's on screen

Your real messages going in as the only source
Four sections drafted: about us, voice and tone, sign-off, standing rules
Style only. Every bullet has to be visible in the samples

Under the hood

Style by contract: no property facts, no door codes, no addresses, no guest names. Facts live in the knowledge base and voice lives here, which is why you can rewrite this freely without worrying you broke the AI's accuracy.

Step 5 of 6 · On rails

The persona sets how it sounds. Policies set what it does

Policies are never free text the model can invent. Each one is a typed rule with typed settings, so the line you read in your dashboard is the line the model is given, word for word.

What's on screen

Stand down while you are replying yourself, silently or as a draft
Always escalate a topic. Never bring one up
A precedence stack the persona cannot climb

Under the hood

The stand-down never reaches the model at all. It runs in tested code before any inference, so there are zero prompt tokens and zero chance the model talks itself out of it. A line reading like ignore previous instructions is declared void, and the rest applies normally.

Step 6 of 6 · Compounds

It gets better every checkout, and corrections stick

A finished conversation is mined once, at the right moment: three days after checkout, and only if the thread has gone quiet for that same window. Any new message restarts the clock, so nothing gets mined mid-argument.

What's on screen

Checkout, then a grace period, then mined exactly once
Every new fact adjudicated against what you already know
A contradiction becomes a reviewable update, old beside new

Under the hood

Most systems that learn continuously are really just accumulating, and six months in they return three contradictory answers to one question. New facts are matched by meaning and resolved at write time. Opposite polarity on the same subject is always a conflict, never a match.

Step 1 of 6 · Connect

It reads the conversations you have already had

Connect your property system and Linkbase does not just pull bookings forward. It reaches backward through your past reservations and pulls the full message thread from each one.

What's on screen

Your PMS connecting, then the backfill walking back through past stays
Every past reservation's message thread pulled and stored
Reservations we already have are skipped without an API call

Under the hood

What it handles

Each of these used to be a task on someone's list. Here's what runs it now.

Writing your own FAQ
Guest
Nothing to write. Linkbase pulls the answers out of threads you already sent and hands them back as a review queue, each one showing the conversation it came from.
Repeat questions
Guest
Wifi, parking, check-in time, the address, how the AC works. Answered from that guest's actual booking and your property guides.
Five platforms at once
Guest
Airbnb, VRBO, Booking.com, direct, email, and SMS all land in one unified inbox.
2 AM messages
Guest
A guest who can't find the lockbox at 2 AM gets a real answer. The AI never goes off shift.
Sounding like a robot
Guest
Your business gets a persona file holding your tone, your phrasing and the sign-off you actually use. We draft it from your own messages, then you edit it like any document.
The same correction, twice
Guest
Tell it why a reply was wrong, once, in plain English. That becomes a standing rule you confirm, and it does not happen again.

Business intelligence

Questions you have never been able to ask

A record of what guests actually ask this business is not something any host has ever had. Once it exists, it answers more than the guest in front of you.

What guests ask that you can't answer
Every fact records that a topic came up in a real conversation. Compare that demand against what you have actually documented and the holes show up by name: parking came up in fourteen conversations across four properties and you have no parking article. Linkbase drafts the missing one.
Which of your rules are really company-wide
When the same thing you said turns up on enough separate properties, checkout is 10 AM on five of six units, it isn't a property rule. It's a company policy, and one account-wide article beats six copies. Genuine exceptions get listed on the suggestion instead of being flattened.
What you changed without updating your docs
Your check-out article says 11 AM and you have told three recent guests 10 AM. That contradiction becomes a reviewable update with old beside new. Approving it rewrites the article and keeps the previous version, so you can always see what it used to say and why it changed.

Every answer came from you, every rule you set it can't climb, and every correction you make it never forgets.

Under the hood

How it actually works

1
It arrives within seconds
Messages come in from your property system by webhook, in whatever channel the guest chose to reach you on, and are matched to the stay as they land.
2
It's matched to the stay before anyone reads it
Each thread is linked to its reservation and property on arrival and scoped to your account, so the guest, the dates, and the full history are already resolved.
3
It reads before it writes
Linkbase pulls the stay itself first, then searches your own knowledge base, and only reaches the open web for things like transit and restaurants. Two lookups is the normal shape of a reply, not an exception. That's how it says “your code is 4827, active from 4 PM” instead of “please contact your host.”
4
You decide what sends on its own
Run review-first while you build trust, then switch to auto-send per property. It has no ability to book, charge, refund or cancel anything, and every reply keeps a trace of the tools it called and the fields it used, so you can check its work.

Connects to

Works with what you run

Your existing tools stay where they are. Linkbase connects to them instead of asking you to move.

See every integration →

Airbnb
VRBO
Booking.com
Email
SMS
WhatsApp

The jobs next to this one

Guest appOne branded guest link for their whole stayRead more→Upsells & revenueOffers made in conversation, paid through StripeRead more→

See it on a real conversation

The live demo runs the same AI on a sample booking. Ask it what your guests ask you.

Try the AI Book a walkthrough

Messaging that never goes off shift.

You have already trained it

It reads the conversations you have already had

What it handles

Writing your own FAQ

Repeat questions

Five platforms at once

2 AM messages

Sounding like a robot

The same correction, twice

Questions you have never been able to ask

What guests ask that you can't answer

Which of your rules are really company-wide

What you changed without updating your docs

How it actually works

It arrives within seconds

It's matched to the stay before anyone reads it

It reads before it writes

You decide what sends on its own

Works with what you run

The jobs next to this one

See it on a real conversation