The problem: using AI today = sending your data to the US
When you use ChatGPT, Claude or Copilot, every question you ask leaves your computer. The text goes to OpenAI’s, Anthropic’s or Microsoft’s servers in the United States. You have no control over what happens to it.
For drafting an email or summarizing a news article, that’s fine. But for a client contract, a financial statement, an HR file or a medical report, it’s a real risk.
- A lawyer pasting a restructuring file into ChatGPT may violate attorney-client privilege.
- A DPO sending personal data creates an undocumented transfer outside the EU.
- A consulting firm running client financials through it risks a data leak it can’t control.
The question isn’t “is AI useful”. It is. The question is: how do you benefit from it without depending on a US provider and without putting your data at risk?
The answer: install everything on your own machines
Today, the same types of AI models as ChatGPT exist as free, open source software. You download them, install them on your own machines, and they run without any internet connection. No cloud. No subscription. No provider seeing your data.
Concretely, this means:
- No data ever leaves your premises. Ever. You can enable airplane mode and the AI keeps working.
- No subscription. The software and models are free and open source.
- No vendor lock-in. You’re not tied to any provider. If a vendor shuts down or raises prices, it doesn’t affect you.
Two options depending on your team size
Option 1 : on each workstation. We install AI directly on your staff’s computers. Everyone has their own local AI assistant. Requires a Mac or PC with at least 16 GB of RAM.
Option 2 : one server for the whole team. We deploy a dedicated machine on your premises (or on your private network) that runs the AI models. Your staff access it from any computer, even a lightweight one. No need to upgrade every desk. The data stays within your walls.
In both cases: zero data on the internet, zero subscription.
Available AI models (all free)
These are open source software, developed by major technology companies:
- Mistral: developed in France. Excellent in French, our default choice for European clients.
- Llama: developed by Meta. Very capable, especially in English.
- Qwen: developed by Alibaba. Excellent at reasoning and analysis.
You can install several, test them, switch anytime. No commitment, no cost.
What your teams can do with it
Here are real examples of what we set up for our clients:
- Analyze a contract: “What are the non-compete clauses? Summarize the conditions precedent.” The AI reads the document and responds, without it ever leaving your network.
- Summarize a 50-page report: you paste the text, the AI produces a structured summary in 2 minutes.
- Process HR data: CV screening, annual review summaries, personnel review preparation. No employee data leaves your network.
- Prepare a financial file: balance sheet analysis, ratio comparison, summary note drafting. The numbers stay with you.
- Write and reformat: meeting minutes, letters, internal memos. The AI drafts, you validate.
For 80% of a professional’s daily tasks, summarizing, writing, analyzing text, local AI does the job just as well as ChatGPT.
For technical teams: a full development agent, running locally
This setup isn’t just for business roles. Your developers can benefit too, and that’s where it gets really powerful.
There’s a tool called Claude Code. It’s an AI agent for developers: it reads a project’s source code, understands the architecture, modifies files, runs tests, automates tasks. It’s the most advanced AI development tool on the market.
What most people don’t know: Claude Code works with any AI model, including free local ones. It’s not limited to Anthropic’s paid models. You connect it to Ollama, choose your model (Mistral, Qwen, Llama…), and your dev team has a full development agent running on your machines. No license fee. No source code sent externally.
A single command:
ollama launch claude
Your developers get the same capabilities as those paying for a cloud subscription, but everything stays on your premises.
What it costs
| Local AI | ChatGPT / Claude (cloud) | |
|---|---|---|
| Software | Free | €20-25 / month / person |
| AI models | Free | Included in subscription |
| Hardware | Your current machines or a dedicated server | Any device |
| Confidentiality | Total, nothing leaves your premises | Data sent to the US |
| Internet required | No | Yes, always |
The software and models are free. If you deploy a dedicated server, the only cost is the hardware (a machine with GPU). We help you with sizing.
Honest limitations
We won’t tell you it’s perfect. Here’s what local AI does less well than the best cloud models (GPT, Claude Opus, Gemini):
- Highly complex reasoning: for 10-step strategic analyses, paid cloud models are still better. For summarizing a contract or drafting a letter, the difference is minimal.
- Speed: on an individual workstation, it depends on the machine. On a dedicated server with GPU, it’s fast.
- Images and audio: most local models only process text today.
These limitations shrink every month. Open source models are advancing very fast. What was impossible a year ago runs smoothly today.
In summary
- It’s free. Open source software, free models, no subscription, no license.
- It’s confidential. No data leaves your premises. Airplane mode, it works.
- It’s for everyone. Business roles (contracts, HR, finance) and developers (full code agent).
- It installs in one day. On individual workstations or on a server for the whole team.
- No vendor lock-in. No cloud provider, no surprise price hikes, no changing terms.
Confidential AI isn’t a futuristic concept. It’s free software you install today on the machines you already have.
Sources
Your data must stay with you? We install your local AI, first call is free →