What is the agentic harness?

The foundation our AI runs on: a connected graph of your guest data, a real set of tools, and the control to run them safely. It is what makes an agent useful rather than a chatbot that guesses.

What can the agent do today?

The CRM agent reads your connected data, runs analysis, builds reports and dashboards you can share, and searches the web. It reads and drafts. Anything that would change your data waits for your sign-off.

Does it work with Square?

Yes. The graph joins what flows through Stampede, including a connected Square till, so the agent reasons over spend alongside bookings, Wi-Fi, reviews and loyalty.

Do I need to set anything up?

No. The graph builds itself from the data already flowing through Stampede. Nothing to export, map or import by hand.

Is this “AI-powered marketing”?

No. We built the foundations first, the connected data and the tools, and the agent is early and growing on top of them. The opposite of a chatbot bolted onto siloed data.

Is my data safe with it?

Graph and database access is read-only, writes wait for your confirmation, and secrets and internal config stay out of reach. It is your own data, on your own account, and yours to export.

Engineering for hospitality

Tools. Context. Control.

The foundations our AI runs on, so it answers from your data and asks before it acts.

See it live

BookingsPaymentsWi-FiReviewsLoyalty

Agent run

Regulars referred by a VIP who haven’t been back?

Tools

Read the guest graph

Score churn risk

Build the report

A win-back list, grouped by the VIP who referred them.

Win-back emailSign-off

writes need your sign-off

Bookings

Wi-Fi

Reviews

Loyalty

Agent run

Regulars referred by a VIP who haven’t been back?

Tools

Read the guest graph

Score churn risk

Build the report

A win-back list, grouped by the VIP who referred them.

Win-back emailSign-off

writes need your sign-off

Powering 2,500+ UK venues

The problem

AI needs foundations.

On siloed data with nothing to do, a chatbot just guesses.

One question, two answers

Before

Bolted on · guessing

ChatbotMade up

How many is Sarah Mitchell’s party on Saturday?

About 12 people, I think!

BookingsNot connected

ActionsNone, it can only chat

A chatbot on tools that don’t talk to each other
It can’t see the booking, so it guesses
It can’t do anything with the answer

After

Answered from her record

AgentYour data

How many is Sarah Mitchell’s party on Saturday?

18 covers, Saturday at 19:30, usually table 12.Nut allergy on her last 6 bookings: one nut-free dish is on the order.

From her guest recordOffer drafted · needs sign-off

An agent on one connected record per guest
It answers from what happened, not a guess
It drafts the next step and waits for your sign-off

Same question, same guest. The difference is what the answer stands on.

How it works

Three parts.

What turns a model into an agent you can trust.

Context, tools, control

01Context

Reads the graph

Every guest, joined up. It answers from what happened, not a guess.

02Tools

Does the work

Queries the graph, reads your data, browses the web, builds the report.

03Control

Asks first

Runs that resume, work in the background, and writes wait for your sign-off.

The graph

All joined up.

Not rows in a table. Your guests, joined by everything they do.

One question, one traversal

ReferredDined withVisitedReviewed

Guest

Venue

Review

Wi-Fi

One traversal

Regulars referred by a VIP who haven’t been back.

Lapsed

Computed

Churn riskLifetime valueInfluence

Referred

Guest

Venue

Review

Wi-Fi

One traversal

Regulars referred by a VIP who haven’t been back.

Lapsed

Computed

Churn riskLifetime valueInfluence

Every guest

Bookings, spend, Wi-Fi, reviews and loyalty on one record.

Who knows who

Who dined with whom, and who referred whom.

Computed signals

Churn risk, lifetime value and influence, worked out.

One hop

One question, one traversal. Not a week of SQL.

Proof

Built, not promised.

40+Node types modelled

20+Signals computed

1Graph across every service

91 releases, logged in the openLatest: Multi-account ordering, 8 Sept

Feature log

FAQ

Questions, answered.

See it working.

The graph and the agent on it, live, on your own data.

See Agent Chat