For businesses
Local AI for your business
Tools, costs and use cases for bringing AI into your company: the data stays on-premises, the costs stay predictable.
Why local AI
The data stays in-house
No document, email or spreadsheet leaves for a cloud server. For a business with sensitive data — clients, invoices, projects — that is the difference between a hypothesis and a decision the board can sign.
Predictable costs
A one-off hardware investment instead of a subscription that grows with every use. After the first year, the infrastructure costs almost nothing but power.
It works offline too
No dependence on connectivity, API limits or provider price changes. The network can be isolated and the model still answers.
What that means in practice
One concrete example: a business that puts its office documents in a searchable archive and generates first drafts without the data leaving the building.
Assistant over internal documents
Summaries, searches and answers over manuals, procedures and the mail archive — with a 7 to 14 billion parameter model.
First drafts of proposals and quotes
Sales starts from a draft generated from the project data instead of a blank page.
Email triage and classification
Incoming requests are tagged by urgency and department before anyone reads them.
First-level customer support
The most frequent answers go out pre-filled from the historical archive; a human signs and sends.
Cataloguing historical archives
Years of unindexed documents become searchable: a weekend batch, not a six-month project.
The full story
How it went in a real business: hardware, model choice, timeline and costs.
What it costs, honestly
| Profile | Indicative hardware | Range |
|---|---|---|
| Trial | 12–16 GB GPU, modern CPU | 1,500 – 3,000 € |
| Light production | 24 GB GPU, 64 GB RAM | 3,000 – 5,000 € |
| Production | 48 GB GPU or multi-GPU | 5,000 – 10,000 € |
Indicative range for a workstation with a 7 to 14 billion parameter model. The setup planner calculates the exact configuration for your case.
The first three steps
- 1
Map the use cases
List the 3 to 5 activities that slow the team down most. No model needed yet: the list is enough.
- 2
Size the hardware
The setup planner turns the list into a configuration: VRAM, RAM, budget.
- 3
Start small
One use case, one team, three months. Then scale — or not, and the investment is still justified.
Ready to bring it in-house?
The guide is free and needs no signup. The setup planner does the maths for you. The newsletter tells you when there is something new.