Sheet 1 of 6 — General arrangement
Depth 000 mm
A private AI system,
drawn before you buy it.
We install AI on hardware you own, then hand over every key, file and document on completion. Below is the plan: what gets built, what it costs, and precisely what — if anything — leaves your network.
Depth 000 mm
- Discovery call
- Free · 20 min
- Pricing
- Fixed, published
- Deployment
- Your hardware
- On completion
- You own all of it
Section through a typical installation
Site boundary — hardware you own
- 01
Local model
OllamaInside — Local compute
The model itself, running on your machine. Answers are generated on your hardware, so a request can be served with no network connection at all.
- 02
Private assistant
OpenClawInside — Automation
The assistant your team talks to, reachable from WhatsApp or Telegram. It holds the conversation and decides which of the runs below to use.
- 03
Document store
Your filesInside — Data at rest
Your documents, indexed for retrieval with citations back to the source file. Stays on your disk. Embeddings can be deleted on request.
- 04
Automation runs
n8nInside — Automation
The workflows that do the repetitive work — intake, extraction, routing, reporting — wired to the tools you already use.
- 05
Commercial API
ChatGPT · Claude · GeminiCrosses your boundary
Optional, and off unless you ask for it. What leaves: the text of each request you route here, which can include passages retrieved from your documents. Nothing is sent on this run until you enable it.
- Depth
- 000 mm — surface
- Scale
- Not to scale
Sheet 2 — Legend
Every class of work, with its price on the drawing
Depth 250 mm
Classification key
Four classes of run appear on this drawing. Each tier below states which of them it puts on your site — so the price and the mechanism are read in the same glance.
- Local compute
- The model and the machine it runs on. Your hardware, your electricity, no network required.
- Automation
- Runs that join two systems together and do the repetitive work between them.
- Data at rest
- Your documents, indexed for retrieval with citations, sitting on your own disk.
- Leaves your network
- An optional route out to a commercial model. Drawn dashed because it is off unless you switch it on.
- BAH-02.1
Personal setup
Individuals and freelancers
Installs
- Local compute
- Automation
- OpenClaw + Ollama installed
- WhatsApp / Telegram integration
- 14-day email support
- BAH-02.2
Business automation
Firms of 5–50 with confidential files
Installs
- Local compute
- Automation
- Data at rest
- Private assistant on your documents
- Up to 5 automated workflows
- Access controls and audit logging
- BAH-02.3
Enterprise systems
Organisations under compliance load
Installs
- Local compute
- Automation
- Data at rest
- Retrieval with source citations
- Role-based access control
- Deletion capability for embeddings
- BAH-02.4
Custom software
Anything beyond AI
Not a classified AI run
- Web and mobile applications
- Full-stack architecture
- Scalable deployment
Retainers run $900, $1,800 or $2,700 a month for 5, 10 or 15 hours, three-month minimum. Productised templates start at $370.
Sheet 3 — Boundary detail
What stays, and what crosses
Depth 450 mm
Most AI vendors ask you to trust a sentence. This is the same claim drawn as a boundary: four components inside it, one optional route out, and a label on that route saying exactly what travels along it.
The default installation has nothing on the crossing at all. It is drawn because pretending commercial models are never the right tool would be its own kind of dishonesty — not because we expect you to use it.
Key to this detail
- Local compute
- The model and the machine it runs on. Your hardware, your electricity, no network required.
- Automation
- Runs that join two systems together and do the repetitive work between them.
- Data at rest
- Your documents, indexed for retrieval with citations, sitting on your own disk.
- Leaves your network
- An optional route out to a commercial model. Drawn dashed because it is off unless you switch it on.
Site boundary — hardware you own
- 01
Local model
OllamaInside — Local compute
The model itself, running on your machine. Answers are generated on your hardware, so a request can be served with no network connection at all.
- 02
Private assistant
OpenClawInside — Automation
The assistant your team talks to, reachable from WhatsApp or Telegram. It holds the conversation and decides which of the runs below to use.
- 03
Document store
Your filesInside — Data at rest
Your documents, indexed for retrieval with citations back to the source file. Stays on your disk. Embeddings can be deleted on request.
- 04
Automation runs
n8nInside — Automation
The workflows that do the repetitive work — intake, extraction, routing, reporting — wired to the tools you already use.
- 05
Commercial API
ChatGPT · Claude · GeminiCrosses your boundary
Optional, and off unless you ask for it. What leaves: the text of each request you route here, which can include passages retrieved from your documents. Nothing is sent on this run until you enable it.
Reference 05 is the only element on this sheet that can send anything off your premises, and it is switched off unless you ask for it.
Sheet 4 — Handover schedule
What you are left holding
Depth 750 mm
Every engagement ends with a handover, and the handover is the product. If we disappeared the week after completion, the schedule below is what would keep your system running without us.
Site boundary on completion — everything inside becomes yours
H1
Source code
Everything written for you, in a repository you own.
H2
Credentials
Every key, login and certificate, transferred to your control.
H3
User and admin guides
Written for your team, not for engineers.
H4
Architecture drawings
The as-built version of the plan on this sheet.
H5
Recorded walkthrough
A video of the system being operated and maintained.
H6
Support window
14, 30 or 60 days by tier — then you are genuinely independent.
Nothing on this schedule is withheld, staged, or held back as leverage. There is no retained credential and no proprietary component you would have to licence from us afterwards.
Sheet 5 — General notes
Read before you commit
- N1
You probably do not need us if a spreadsheet, a shared inbox rule, or a single template would solve the problem. We will say so on the call and bill nothing for saying it.
- N2
Every price on this sheet is fixed for the scope stated and published before you commit. No hourly billing. Scope changes go through a written change request, or they do not happen.
Legend — sheet 2
- N3
The default installation sends nothing to a third-party model. Where a commercial API is genuinely the right tool, it is drawn as a boundary crossing (05) and switched on only at your instruction.
Detail 05 — boundary crossing
- N4
Bespoke AI Hub Ltd is registered in England and Wales and works remotely with clients worldwide. Personal data is handled under the UK GDPR.
- N5
This drawing is illustrative of a typical installation. Yours is drawn to your own requirements during the discovery call, before any work is agreed.
General arrangement — BAH-01/1
Sheet 6 — Issue
Book the call. If AI is not
the answer, we will say so.
Twenty minutes, free, no obligation. We ask what is eating your team's time, tell you honestly whether this is worth building, and if it is, put a fixed price on it in writing.
Scheduling runs through Calendly, so your name and email reach them too. Email us directly at hello@bespokeaihub.com if you would rather it did not.
- Duration
- 20 minutes
- Fee
- Free
- Obligation
- None
- Follow-up
- Fixed quote in writing
- Deeper session
- $140 · 60 min
- Response
- 1 business day