AI for your business
Put AI to work in your business, with the checks it needs
We build AI into the software your business runs on: assistants that answer customers and staff from your own content, tools that draft listings, recommendations and write-ups from your data, and camera analysis that runs on the device. The rule we build by: your code decides the prices, scores and numbers, the AI writes the words around them, and its work is checked before anyone sees it. Every example below is working code you can try. The first call is free.
Assistants · drafting · lead tools · camerasConnected to what you already run
- Your rules decide, the AI writes: prices, scores and every number come from your code, and a reply with a number we didn't give it never reaches a customer
- Answers from your own content: assistants built to say “I don't know” rather than guess, even when someone tries to trip them up
- Working examples, not slides: every AI feature on this page runs on this site today; try them before you call
Tell us what you want AI to do Try a working example
The first 30-minute call is free. We usually reply within one business day.
AI working on this site now
- Ask AI Jan A chat assistant that answers visitors from this site's own pages Says “I don't know” rather than guess; the AI key stays on the server and a daily cap keeps the bill predictable.
- Keelstone IT An AI plan finder, IT health check and ROI calculator that qualify leads Code decides the plan, the scores and every number; the AI writes the words, and a reply with a number we didn't give it is replaced.
- Gavelry An AI listing writer: a seller's sentence in, a full auction listing out Prices come from sales data, never from the model; each draft is checked and trimmed before it's shown.
- Passerby Pedestrians, bikes and cars counted on traffic cameras and any video Detection runs in the visitor's browser; their own video is never uploaded.
- Solenne Plastic Surgery Live face tracking: visitors preview a procedure on their own face Tracking and the preview run only on the visitor's device.
- Virtual Salesman A 2006 Delphi program moved to the web by an AI coding agent in about an hour Its rules written down first; the author's sample documents match to the bit and the totals to the cent.
Jobs AI does well when it's built in properly
Answering the same twenty questions
Hours, policies, prices, how to book: an assistant on your site or intranet answers from what you've already written, at any hour.
Drafting from your facts
Listings, plan recommendations and write-ups for finance, drafted from your data in seconds, with the numbers checked against it.
Qualifying leads
A plan finder or a health check that asks a few questions, scores the answers by your rules and explains the result in plain words.
Seeing with a camera
Counting people, bikes and cars, or tracking a face for a live preview, on the device, without sending video anywhere.
Building your software faster
AI coding agents with written rules and every line reviewed: a desktop program moved to the web in about an hour, a platform rewrite tracking about four times faster.
Connecting it to your systems
Through APIs to your database, CRM and the services you already use, so the AI works with your real records, not copies.
What goes wrong with AI, and what we build against it
| The risk | What we build |
|---|---|
| It makes up numbers | Prices, scores and amounts come from your code. Every amount and percentage in a reply is checked against the facts it was given; if one doesn't match, a plain template is shown instead. |
| It makes up answers | It answers only from your content and says “I don't know” otherwise. It's tested against real questions before launch, the same way you'd test any software. |
| Someone tells it to ignore its instructions | Strict written rules, a model that only ever writes text, and checks on what it writes before it's shown. |
| The bill runs away | A low-cost model where it's enough, and a daily cap that keeps the cost predictable even if the site gets busy or someone tries to abuse it. |
| Private data leaves the building | Camera and photo work runs on the visitor's device. Your AI key stays on your server, and you choose a provider that fits your privacy obligations. |
| You're locked into one vendor | The same code runs on DeepSeek, Anthropic's Claude or any OpenAI-compatible API, switched by a setting. |
Five steps, and you see each one
- A free 30-minute callWhat you want done, who does it today, what it runs on and where the data lives. No preparation needed.
- A first look at the workWhich tasks the AI takes on, what data it may see, which rules must never be left to a model, and what happens when it's wrong.
- A written proposalScope, approach, timeline and a price range, so you know what you're getting before any work begins.
- Built and tested in small stepsYou try each part as it's done. The AI is tested against your real questions and documents, and people trying to break it.
- Launch and handoverDeployment, documentation and a walkthrough, so your team can run it, or we keep running it for you.
The questions that decide whether AI is the right tool
Where the data lives
Which systems hold it, who may see it, and whether it can leave your network at all.
What must stay with people
Approvals, prices and anything with legal weight are decided by rules or by a person. The AI drafts; it doesn't sign.
What a wrong answer costs
A wrong opening hour is a nuisance; a wrong dose or a wrong price isn't. The checks are sized to the cost of a mistake.
Volume and running cost
How many questions or documents a day, so the model and the cap fit the budget before launch.
Frequently asked questions
Which AI model do you use?
The one that fits the job, the budget and your privacy obligations. Our assistants run on DeepSeek, Anthropic's Claude or any OpenAI-compatible API, switched by a setting, so you're not locked in to one vendor.
Will it make things up?
A language model can, so we don't let it decide anything that matters. Your code supplies the prices, scores and numbers; the model writes the words around them, and its reply is checked before it's shown. Assistants answer only from your content and say “I don't know” otherwise.
Do we have to replace our systems?
No. AI is added to what you already run, through APIs to your database, CRM and the services you use.
What does it cost to run?
It depends on the volume and the model. For a website assistant on a low-cost model, each question costs a fraction of a cent, so a few hundred questions a month typically comes to a few dollars or less. A daily cap keeps the bill predictable.
What does a project cost?
The first 30-minute call is free. After a first look you get a written proposal with the scope, the approach, a timeline and a price range.
Can you help our own developers use AI?
Yes. As a fractional CTO we set up AI coding tools for your team: project rules, how work is split into small pieces and how every change is reviewed.
Tell us what you want AI to do
A few lines is enough to start. If you know them, include:
- The job you want AI to take on, and who does it today
- What it would need to read: your pages, documents, a database
- What must never be left to the AI: prices, approvals, anything regulated
- Roughly how often it happens: a day, a week, a month
I read every message myself and usually reply within one business day. Prefer more detail? Fill in the project questionnaire (about 15 minutes).