A private AI demonstration · Chatting across the systems you already run
Chat across your systems
One secure, private AI assistant, answering across your CRM, your finance system, your spreadsheets and your documents at once.
Your answer is rarely in one place. It sits split across a CRM, a finance system, a spreadsheet someone keeps by hand, and a folder of documents, none of them built to talk to each other. This is one private AI assistant that reads across all of it and answers in plain English.
It runs on a made-up charity, the Fenmere Trust, so I can show you every screen without exposing anyone's real data. The systems it connects to are real.
How it works, and why that matters
The assistant runs on open weights in a private account. Your questions and its answers are never sent to ChatGPT, Gemini, Claude or any other outside service. What makes it different from a public tool is not just where it runs, but what it can reach.
It connects to the systems the charity already uses: its CiviCRM database, its Xero finance system, a Google Sheet the volunteers keep, and a set of Google Docs. One question can pull from all four at once and join the answer for you.
And every figure it gives you is a link. Click a number and it opens the exact CiviCRM record, the Xero transaction, the spreadsheet row or the paragraph in the document it came from. Nothing is taken on trust, and nothing left the account it runs in.
Meet the Fenmere Trust
The Fenmere Trust is a small, invented conservation charity in the Norfolk Breck. It manages a handful of nature reserves, protects scarce Breckland wildlife, and runs a youth rangers group for local teenagers. It is fictional on purpose, so I can open every screen for you rather than hide behind a real client's confidentiality.
Like most charities, its information is scattered: members and donations live in CiviCRM, the accounts in Xero, the volunteers' wildlife counts in a Google Sheet, and its policies, case notes and grant agreements in Google Docs. The examples below each start with a real job, then show the question you would ask and the answer that comes back.
Reconcile across two systems
The job every finance volunteer dreads: the CRM and the accounts don't quite agree, and finding out why means an afternoon of cross-checking two systems by hand.
You ran a spring appeal, and you want to know it has all landed properly before you report to the board.
You'd ask:
"For the Fenmere Wetland Appeal 2026, what did the CRM and the finance system each record, and if they differ, why?"
Protect what has to stay secret
Some of a charity's data is genuinely dangerous to share. The assistant answers from it, but knows the rules and refuses to hand over what shouldn't leave.
A local group asks what you've recorded about a sensitive species. You want to give them the figures, but hold back the protected location.
You'd ask:
"A local birding group has asked about stone-curlew at Fenmere. What do our surveys show this year, and what are we allowed to tell them about the location?"
Pull a board briefing together in one go
The kind of question that usually means opening three or four systems and stitching the answer together by hand: the headline position a trustees' meeting actually needs.
A trustees' meeting is coming up and you want the charity's headline position in one go, rather than opening every system in turn.
You'd ask:
"Prepare a short briefing for Thursday's trustees: current membership numbers, the wetland appeal total, the Heritage Fund grant position, and anything overdue."
Understand your membership
The questions your CRM makes hard: retention across a whole membership, and the quiet data problems that skew every report until someone finds them.
You're worried about retention and want to see where you're losing people, without waiting on a custom report.
You'd ask:
"Which members lapsed this year, and which regions are we losing fastest?"
Before you trust any of the numbers, you want the messy data cleaned up: the same supporter entered twice under slightly different spellings.
You'd ask:
"Are there any duplicate members we should merge?"
Your systems will be different. That's fine.
The Fenmere Trust runs CiviCRM, Xero, Google Sheets and Google Docs because those are common and I can show them openly. But the four systems here are examples, not a list of what's supported.
The honest rule is this: if a system has a way in, it can join the assistant. That covers anything with an API (most modern CRMs like Beacon, Donorfy and Salesforce, finance systems like Xero, QuickBooks and Sage, Microsoft 365, Google Workspace, Airtable), anything with a database underneath it, and anything that can produce a regular export, which is nearly everything else. Each system is connected once, read-only, and from then on it's just another place the assistant can look.
So the question isn't whether your stack is on a list. It's which of your systems hold the answers your team keeps stitching together by hand.
The questions charities ask first
How do you stop it making things up?
AI tools are known for inventing plausible answers, what the industry calls hallucination, so this is the right first question. Three things keep this assistant honest. It never does the arithmetic: every count and total is worked out by ordinary code reading your systems, and the AI's job is to phrase the answer, not calculate it. Every answer has to cite its source, and the link is built from the record itself, so it cannot invent one. And before anything reaches your team it is tested against data where the right answers are already known, every question asked again and again, with a single wrong answer counting as a fail. When it cannot stand a claim up against your data, it tells you so rather than guessing. There is a fuller account of how I test this, and what the testing has caught, on this page.
Can it change or delete anything in our systems?
No. The assistant connects to each system with its own limited, read only access. It can look records up; it cannot edit, add or delete them. If it tried, your systems would refuse.
Where does our data actually go?
Nowhere. The assistant runs in a private account, your questions and records are never sent to ChatGPT, Gemini, Claude or any other outside service, and the answers stay between you and it. That's what makes it safe to ask about a safeguarding note or a donor's giving history.
What does it cost, and where would we start?
It's a custom build, so it depends on your systems and where the pain is. Most charities start small: one system, one job that's hurting, proven properly, then widened out. Email me and tell me what your data is scattered across, and I'll tell you honestly whether this fits.
This is a demonstration, not a product
There's nothing to sign up for here. The Fenmere Trust isn't real, and its systems are only running while I film. What's real is the build itself: this is the kind of private AI assistant I make for UK charities, connected to the systems you already run, answering across all of them, with nothing going to a public AI.
If you'd like to see what this would look like over your own CRM, accounts and documents, email me and tell me what your data is scattered across.
The Fenmere Trust is a fictional charity and all of its records are synthetic. The systems the assistant connects to are real: a self-hosted CiviCRM, a Xero demo company, and Google Workspace, all in an account I control.
