On-Premise & Local AI
AI that runs where your data already lives.
For legal, healthcare, finance, defence-adjacent and IP-heavy teams, sending data to a third-party API is often not an option. We deploy open-weight language and vision models on your hardware, in your private cloud or fully air-gapped, and build the applications on top.
Schedule 1
What is included
- B-3.1
Hardware and model sizing
Which model, at what quantisation, on which GPUs or CPUs, for your users and your latency target. You get a written bill of materials before you buy anything.
- B-3.2
Private model serving
Production inference with vLLM, llama.cpp or Ollama behind your identity provider, with quotas, logging and an OpenAI-compatible API your tools already speak.
- B-3.3
Private RAG and assistants
Document search and chat over your file shares and systems, with permissions that mirror who can already read what.
- B-3.4
Air-gapped and edge deployment
Offline installers, signed model artefacts and update procedures for networks with no internet at all, and for devices in the field.
- B-3.5
Operations handover
Monitoring, upgrade runbooks and model-swap procedures your IT team can run without us.
Schedule 2
Typical projects
- Law firms searching privileged documents
- Clinics summarising notes without sending patient data out
- Manufacturers keeping process IP inside the plant
- Public-sector and regulated environments
- Developer copilots that never send source code off-site
Materials
- Llama
- Mistral
- Qwen
- Gemma
- vLLM
- llama.cpp
- Ollama
- NVIDIA
- Kubernetes
- Open WebUI
Schedule 3
How it runs
- 01
Discovery · 1–2 weeks
We map the problem, the users, the data and the constraints, and agree what success will be measured by.
- 02
Blueprint · end of discovery
Architecture, scope, milestones and a fixed price for the first build phase, all in writing. You can stop here and keep the blueprint.
- 03
Build · weekly demos
Short iterations with a working demo each week, automated tests from day one and a shared board you can always see.
- 04
Launch & handover · planned cut-over
Staged rollout, monitoring switched on, documentation and a recorded handover session with your team.
- 05
Run & improve · optional
A monthly support plan, or nothing: the handover is written so that you never need us to keep the lights on.
Schedule 4
Common questions
Are local models good enough?
For many focused business tasks (search, summarising, extraction, drafting) current open-weight models are close to hosted ones, especially with good retrieval. We benchmark on your own tasks before you commit to hardware.
What hardware do we need?
From a single workstation GPU for a small team to a multi-GPU server for organisation-wide use. Sizing is the first deliverable, and it depends on model size, users and response-time targets.
Can it work with no internet connection at all?
Yes. We have procedures for fully air-gapped installs, including how models and updates cross the gap.