Ask your data
“Which clients had no orders last month?” The answer is built from the same functions that draw your existing screens, so the number matches the screen, with a link to it.
Ask your business a question, get an answer with a source
An assistant is only useful if you can trust the answer. Ours answer from your own systems and documents, show the source for every number, and can’t change anything unless you decide they should.
Three kinds of assistant
Ask your documents
Policies, contracts, manuals and past proposals. Answers quote the passage and link to the page.
Connect the AI you already use
Read-only connectors (MCP) let Claude, ChatGPT or Microsoft Copilot read your CRM or database without being able to change it.
How we make answers trustworthy
Tested tools, not free-form queries
The AI never writes free-form database queries. It calls a small set of tools we write and test, and every answer carries its source.
Scope fixed on the server
Each customer’s scope is fixed on the server, not chosen by the AI, so one customer’s assistant is kept out of another’s data, and we test for it.
Personal details kept back
Personal details are removed before they reach a model, where they aren’t needed for the answer.
Tested against known answers
We test against questions with known answers, including a decoy dataset designed to catch leaks between customers. It’s built and tested to say when it can’t find an answer, rather than guess.
Live
A warehouse analytics assistant
For a warehouse analytics platform used by US third-party logistics companies, we built an assistant that answers questions like “which clients had no picking last month?”. It can only call read tools that wrap the functions behind named screens, so every number matches a screen and links to it. It’s tested against a dataset with known answers, plus a decoy tenant that reuses the same client codes.
Our own CRM, connected to AI safely
Tonic Desk, the CRM we build, ships a connector that gives AI assistants 19 read tools and no write tools. Whatever an assistant is asked to do, it can’t change or delete a record.
You choose where the AI runs
Tier 1
A hosted model, through a UK or EU endpoint
The best quality for the least effort. Before any of your data goes to a provider, we check that its business terms rule out training on it, and we tell you which region does the processing.
Tier 2
Inside your own cloud account
The model runs in your AWS or Azure account, under your access controls and your logs.
Tier 3
A private model on hardware in the UK
Open-weight models on your hardware or ours. Nothing goes to an outside AI provider. Our own research and document reading run this way.
Storage location and processing location aren’t the same thing. We’ll tell you both, in writing.
Questions
Is this a chatbot?
It’s an assistant for your team, inside your systems. It can have a chat box, but the point is the answer and its source.
Can it use the AI we already have?
Yes. We can build connectors for Claude, ChatGPT or Microsoft Copilot.
What if it gets a number wrong?
Because every answer links to its source, a wrong answer is easy to spot. We track accuracy against the test set every time something changes.
Get in touch
Tell us the question you’d ask first.
One paragraph is enough. We’ll reply with honest thoughts on whether we can help, and what it would cost.