Hi, it’s Dawn. My laptop won’t turn on since the weekend. There’s a little light but the screen stays black. I work from home so I’m a bit stuck. Any chance someone could look this week?
How the AI behind it works.
No mystery, and no one else’s cloud. Here is what happens to a message, the hardware that reads it, and what is and isn’t kept.
One message, start to finish.
Most of the job is plain, fixed steps. The AI does one part: reading and drafting. You check it before anything goes out.
- Fixed rule
It lands in the in-tray
A message arrives on WhatsApp or email and is copied into the tool. No AI involved.
- The AI
Read and drafted, here in town
Our AI in Bishop’s Stortford reads it and writes a reply. It never promises a price or a day.
Hi Dawn, sorry to hear your laptop is playing up. That sounds like a common issue we can often resolve. We’d be happy to take a look. Please send it in on [day] and we’ll assess it then. Let us know if that works for you. Stort Valley ComputersWritten in 0.83s on 9 November 2026. Real output, unedited.
- You
You check it
Fill in the day, change a word, or bin it. Nothing is sent until you say so.
- Fixed rule
Sent, and logged
The reply goes out and the job is written into your list, so nothing is forgotten.
Where your message goes.
From your phone, to our server, to the AI machine, and back. All of it in Bishop’s Stortford.
- Not sent to OpenAI, Google or anyone else.
- What you paste into the demos isn’t saved or used for training.
- In a tool we build you, what gets kept, and for how long, is agreed with you first.
Your phoneor shop PC
Our serverBishop’s Stortford
The hardware.
Built by us for running AI, in Bishop’s Stortford. It’s what the demos on this site run on.
What’s in it
- Graphics
- 2 × NVIDIA GeForce RTX 3090, 24GB each
- Processor
- AMD Ryzen 9 9950X3D, 16 cores
- Memory
- 64GB
- Model
- Qwen 3.8, 27 billion parameters (open weights), 8-bit
- Speed
- First word in 0.15s, a short reply in 1.09s
Mon Nov 9 14:01:06 2026 GPU Name Temp Power Memory (MiB) 0 GeForce RTX 3090 35C 6W / 300W 23011 / 24576 1 GeForce RTX 3090 34C 12W / 300W 23365 / 24576 Processes: 0 VLLM::Worker_TP0 22988 1 VLLM::Worker_TP1 23008
$ python3 chat.py › "Hi, it's Dawn. My laptop won't turn on since the weekend…" Hi Dawn, Thanks for getting in touch. I’m sorry to hear your laptop is playing up… … bring it in on [day] for a quick check-up. -- first word 0.15s, done 1.09s GPU 0 44C 270W 99% GPU 1 45C 251W 100%
What is and isn’t kept.
| Where | What happens to your text |
|---|---|
| The demos on this site | Read by the AI, answered, then gone. Not saved, not logged, not used for training. Only a count of tries is held for an hour, to keep the demos fair for everyone. |
| A tool we build you | Whatever the job needs, agreed with you before anything is built: for example, an enquiry log you can look back through. Nothing beyond that. |
| Rent our compute | Your requests are answered and not stored or used for training. |
| Your own machine | It all stays on your premises. We only see it if you ask us to help. |
Questions people ask.
Is my data sent to ChatGPT, Google or anyone else?
No. The demos and the tools we host run on our own hardware in Bishop’s Stortford. Your text goes from your browser to our server and to the AI machine over our own network, and nowhere else.
Which AI model is it?
An open model from the Qwen family, with 27 billion parameters, running 8-bit on two RTX 3090 graphics cards. Open means anyone can download and run it; nothing about it is secret, which is part of the point.
Can it get things wrong?
Yes. That’s why our tools draft rather than send, use words like “likely” instead of “definitely”, and are told never to invent a name, price or date. You stay in charge of what goes out.
What if the AI can’t answer straight away?
Every tool we build is designed for it. Depending on the job it waits and tries again, follows a simple rule, or hands over to a backup, so a customer is never left without a reply. We agree which before anything is built.
Do I need new software or a new computer?
Usually not. Most tools work in a web browser or plug into what you already use, like Gmail, Outlook, Google Sheets or WhatsApp. Running AI on your own premises does need a capable PC, which we can build.
Does it learn from what I type?
No. It doesn’t keep what people type. It gets better when we improve its written instructions.