Pricing
What an AI system costs depends on what it has to do: the systems it connects to and the documents behind it. We price each project against its requirements, in a fixed quote agreed before any code.
The call is free. The quote is fixed, in writing and net of VAT, before the build starts.
Keeps it lean: One tool, read-only Adds to it: Several systems it reads from and writes to
Keeps it lean: A tidy knowledge base Adds to it: Large or messy archives, scans and PDFs
Keeps it lean: An existing model, by API or open-weight Adds to it: A model trained for your task
Keeps it lean: One Adds to it: Several
Your price A fixed quote, agreed after the call
Whatever the scope, none of this is an extra. Not hours: a working system, and everything it needs to stay working.
We load your knowledge base, documents or product data and pick the retrieval mode your content needs. Every answer cites its source. When the answer isn’t there, it searches again or says it doesn’t know.
A commercial API model or an open-weight one, large or small, chosen for your budget and your data rules. Switching to a better or cheaper one later is a configuration change, not a rebuild.
Integration with the tools it needs to read from or write to, such as your CRM, product catalogue or helpdesk, through their API or a webhook. The exact scope is written into the quote.
Your real questions become a test suite. We agree the pass mark with you before the build starts, and nothing goes live until it is met.
We deploy it where your data rules point — hosting we manage, your own cloud, or your servers — with monitoring from day one, plus a dashboard showing what each request costs, with limits you set.
Every assistant tells users it is an AI from the first message. From 2 August 2026, Article 50 requires that people are told they are dealing with an AI system, unless that is obvious.
Pay as it proves itself. You never pay in full for something that doesn’t work.
You describe the problem; we tell you whether it needs AI at all, and roughly what it would take.
Nothing to pay
Scope and price, set to your requirements, agreed in writing before we start. No open-ended day rates. That price holds for that scope: if something material changes later, we re-quote it in writing before the work starts.
Nothing to pay
We assemble the core, integrate your systems and test against your data. You see a working demo on your own documents before launch.
We monitor and improve it; you use it, and hosting sits with us or with you, whichever your data rules prefer. Core updates and model upgrades reach you without a new project. The monthly fee is set in the same quote.
After launch there are two running costs: a monthly fee for monitoring and updates, and the model usage your assistant generates. Hosting is your call — we can run it, or you host it yourself and pay for the rest only. Both are set out in the quote, and the cost dashboard shows usage per request, with limits you set.
Our first five projects get 35% off in exchange for a public case study with real numbers and a reference call.
It depends on the project and its requirements: how many systems it connects to, the documents it answers from, whether it needs a custom-trained model and how many languages it serves. After a free 30-minute call you get a fixed quote in writing, net of VAT, before any code is written.
Two projects with the same name can need very different amounts of work. A list price would overcharge the simple ones and surprise the complex ones later. A fixed quote per project does neither.
Integrations with several systems, a large or messy document base, on-premise deployment, or a task that needs a custom-trained model. We flag on the call which of these apply to you. Pricing them and setting a timeline takes us a little longer — both come back to you in writing, before you commit to anything.
The core we build on stays ours; the business logic you ordered is yours — the way your data is collected, processed and turned into a result. Generic building blocks don’t count: a class that writes files to Google Drive stays ours, a class that processes your data your way is yours. Any model trained specifically for you is yours too. If you want the code in-house, you can buy a licence to use the core in your own projects, with credit to QuickScratch. You can then step away from our support and carry on alone — we just need that confirmed in writing.
You don’t have to take it on faith. Payment is split into three milestones, the last one after acceptance tests agreed before the build starts, so you never pay in full for something that doesn’t work.
How it actually starts. With an email, not a form. Write to contact@quickscratch.io.
A 30-minute call. Bring one real process or one sample document. We’ll tell you whether it needs an LLM, a small model, or just a script, and what it would take. No deck.