280×
Running AI got 280 times cheaper
Using a model at GPT-3.5 level became more than 280 times cheaper in two years: from $20 to $0.07 per million tokens.
Some problems have no package that solves them. Then we build one: on your own data, connected to your systems and designed with the people who will use it.
Why now
280×
Using a model at GPT-3.5 level became more than 280 times cheaper in two years: from $20 to $0.07 per million tokens.
1.7%
On the Chatbot Arena benchmark, the gap between the best closed AI model and the best open model shrank from 8% to 1.7% in a single year. A private model on your own servers is now a realistic option.
47%
Of Dutch companies that considered AI but did not start, 47% had concerns about data protection and privacy. A model on your own servers addresses exactly that.
What we build
We start from your problem, not from a technology. The result can be small and specific, as long as it saves time every day.
Knowledge
Manuals, procedures, contracts and past projects in one place where colleagues can ask questions. Every answer shows the source it came from.
Tools
A planning tool, a calculation module or a dashboard that brings figures from several systems together. Built for how your team works.
Portals
A portal where customers check orders, request a quote or get advice, with AI doing the work behind the scenes.
Data
Product data, customer records and spreadsheets enriched, cleaned and connected, so the rest of your software can rely on them.
Websites
We build websites the same way we build software: with AI, on your own content and around what your customers want to do. A first version is there in weeks rather than months, and you see everything before it goes live.
Already have a website that should be found more often? That is AI marketing.
Private AI
Not every company wants its data to pass through an American cloud. We set up an AI model that runs entirely in your own environment, on your own server or in a private European cloud. It is trained on the language of your sector and works for your team only.
A general model knows a little about everything. Yours knows your sector inside out.
An open-source model such as Llama, Mistral or Qwen, sized for your task and your hardware.
We fine-tune it on your documents, terminology and examples, so it writes and reasons the way your team does.
On your own server or in a private European cloud, connected to the systems your team already uses.
We monitor the quality of the answers and train the model again when your work or the rules change.
Not always the best choice: for broad, general questions a large cloud model is often stronger. We tell you honestly which fits your situation.
From idea to software your team uses every day, in steps you can stop after.
We look at the problem, the data and the people who'll use it, and tell you honestly whether it's worth building.
A first version on your own data that your team can try. You see what it does before you invest further.
We turn the prototype into a product: secure, connected to your systems and tested with real users.
We host, monitor and improve it for a fixed monthly fee, and add features when your work changes.
Hosted on European servers, under a data processing agreement. Your data is never used to train language models.
We build with AI ourselves, so a first version takes weeks rather than months. Ownership, hosting and maintenance are agreed on paper before we start. Everything is documented, so another party can take over if you ever want that.
AI product development is building software for problems that no off-the-shelf package solves, on your own data and connected to the systems you already use. One Less Thing builds knowledge assistants that answer from your own documents and show their source, internal tools and dashboards, portals for your customers, and websites built with AI that Google and AI search engines can read. It works in steps you can stop after: an idea check, then a working prototype on your own data within two to four weeks at a fixed price, and only then the full build. Everything is hosted on European servers under a data processing agreement, and your data is never used to train language models. If your data may not leave the company, we set up an open-source model on your own server or in a private European cloud. Ownership, hosting and maintenance are agreed on paper before we start.
Often that is the right answer, and then we will say so. Custom software makes sense when the work depends on your own data, has to fit into your systems, or has to run without someone typing prompts all day.
That depends on what it needs to do. The first conversation is free of obligation and a prototype has a fixed price. Only once the prototype has proven itself do we price the full build, on paper and before we start.
A first prototype is usually ready within two to four weeks. How long the full build takes depends on the connections and the number of users. You will know before we start.
Yes, as part of product development. We build your website with AI, on your own content: fast, readable for Google and AI search engines, and connected to your CRM or inbox where that helps. A first version takes weeks rather than months. Want an existing site to be found more often? That is AI marketing.
We agree that on paper before we start, including what happens to the code and the data if we stop working together. We document what we build, so another party can take it over.
On European servers, under a data processing agreement. If you have stricter requirements, for example your own cloud environment, we build it there.
Yes. We can set up an open-source model that runs entirely in your own environment, on your own server or in a private European cloud. Your data then never leaves your company. It does need suitable hardware or a private cloud server, and you hear up front what that costs.
Fine-tuning means training an existing AI model further on examples from your own sector: your documents, your terminology and the way you write. The model then knows your jargon and follows your templates without long instructions. For one clearly defined task, a smaller trained model often performs as well as a large general one.
Tell us in a few sentences what you're looking for, or which work comes back every week. You'll hear from us within one working day.