Secure AI, dedicated to your company Train. Build. Save.
KastorLab sets up AI in your company: on a dedicated instance in Europe, or on your own infrastructure, with network-isolated agents and the models of your choice, local or external. We train your teams, build your business agents and run it all for you, at a controlled cost per task.
- Accounting
- Finance
- Audit
- HR and payroll
- Legal
- Compliance
- IT
- in internal mode: no external model, no web search
- Closed loop
- in internal mode: no external model, no web search
- one isolated stack per company, in Europe or on your premises
- Dedicated instance
- one isolated stack per company, in Europe or on your premises
- isolated agents, flagged requests, human approval
- Secure AI
- isolated agents, flagged requests, human approval
- per run of a program agent
- 0 tokens
- per run of a program agent
Every model, one place
The best of every AI, at the right price.
Our ambition: the performance of the best models, at a lower cost. KastorLab brings together the market's leading models and open-weight models hosted in Europe or installed on your premises. For each task, Nestor picks the right model, so everyday work no longer pays premium-model prices.
- Sensitive dataOpen-weight models hosted in Europe or installed on your premises, so this data is not sent to an external provider.
- Complex tasksThe most powerful models, only if you allow them. Every outbound request is flagged.
- Repetitive tasksA program agent, tested on your files, that runs with no model: 0 tokens.
- Open-weight models (Europe or on your premises)
- If you allow them: external models
Mistral, Llama, Qwen, Gemma, DeepSeek, Granite, gpt-oss, Claude, GPT, Gemini
Illustration. Logos belong to their owners; their presence implies no partnership.
For every team
Agents for every team.
Every department has repetitive work on files. Pick your team.
- Accounting Entries, VAT checks, year-end meetings.
- Finance Cash flow, forecasts, intercompany.
- Audit and statutory audit Ledger tests, analytical review, notes.
- HR and payroll DSN checks, payroll variables, termination pay.
- Legal Contracts, shareholders' agreements, AGMs.
- Compliance and risk KYC files, watch lists.
- Sales Account scoring, pipeline, follow-ups.
- IT and AI governance Track and reduce token spend.
- Operations and assistants Standard emails, reminders, automation.
Our approach
Train, develop, save.
-
Train
The beaver passes on its know-how.
On-site workshops on Claude, ChatGPT, Gemini and Nestor: know what to ask AI, and what not to hand over to it.
See the training -
Develop
The beaver builds.
Agents that work on your files, with a human sign-off before each one goes live.
See the agents -
Save
The dam wastes nothing.
A managed AI toolbox, powered by Nestor, that lowers the cost per task.
Discover Nestor
Security
Your data stays under your control.
From storage to use, you decide what may leave, and you see everything that does.
- Secure AI
- Closed loop
- Dedicated instance
- GDPR and AI Act preparation
-
Dedicated instance
An instance dedicated to your company, in the European Union, or on your infrastructure with the Sovereign offer.
-
Closed loop
Agents run in an isolated, read-only container: nothing gets in or out while they run.
-
Internal-only mode
No external model, no web search: your questions are handled by your local models.
-
Automatic masking
Some personal data (emails, IBANs, numbers…) is masked before any text is sent.
-
Flagged outbound requests
If an external model is allowed, every outbound request is logged and flagged in the conversation.
-
Human approval
Every agent is approved by someone other than its author; a sealed audit pack records versions and approvals.
The team
The people behind KastorLab.
To write to them, use the company's contact address.
-
Yaniv Benichou
-
Betsalel Elgrably
-
David Dahan
FAQ
Your questions, answered.
What is KastorLab?
KastorLab sets up secure AI in companies: on a dedicated instance in Europe, or on their own infrastructure for the most demanding (defence, banks…), with closed-loop agents. We train the teams, build the business agents and run it all. Our ambition: the performance of the best models, at a lower cost. Nestor powers our toolbox.
Which AI models can I use?
Open-weight models hosted in Europe or installed on your premises (Mistral, Llama, Qwen, Gemma, DeepSeek, Granite, gpt-oss families), chosen to fit your hardware. If your organisation allows it, external models: Claude, GPT, Gemini. For each task, Nestor picks the least expensive model capable of it.
Is my data sent anywhere?
In a closed loop ("internal only" mode), no external model is called: agents are built and run on your company's instance, with open-weight models. If you allow an external model, you leave the closed loop for those uses: to build an agent, your request and the structure of your files go to the model, never your actual rows; in the conversation, each message is sent with the history, the retrieved team memories and the text of attached files, after automatic masking of some personal data. These requests are flagged and logged. Agents themselves always run in an isolated container with no network access.
What about the GDPR and the AI Act?
We claim no certification. We prepare compliance with you: documentation, logs, an audit pack per agent, and an "internal only" mode for the most sensitive data.
What is the AI Act, and what should I do?
The AI Act is the EU regulation on artificial intelligence, applying in stages. Among other things, it asks providers and organisations that use AI to take measures to support their staff's AI literacy (Article 4, current text in force since 27 July 2026). We offer training and an audit (inventory of your uses, classification, gaps, action plan) to help you prepare. We issue no certification and give no legal advice.
Where is Nestor hosted?
Two modes. An instance dedicated to your company, in the European Union, run by KastorLab (Team and Business offers). Or a deployment on your servers or your cloud, with your own storage (Sovereign offer), for defence, banks and organisations that want to keep their data highly confidential.
What hardware do I need?
Nestor installs on a Linux server (Debian 12 or Ubuntu 24.04) or a workstation (macOS, Linux, or Windows via WSL 2 with Docker Desktop). Docker is required: it isolates agent runs. To run models locally in good conditions, plan for a graphics card with at least 12 GB of video memory; on an Apple Silicon Mac, 16 GB of unified memory or more. Agents that are already built need no model at all.
How much does it cost?
Indicative prices are on the Offers page; they are set with our first client companies. A program agent run uses no tokens; with a local model, tokens aren't billed.
Why use Nestor instead of asking Claude or ChatGPT for a script?
Writing a script once is easy; keeping it correct, checked and private over time is harder. In Nestor, the AI writes the agent from the structure of your files, never from your real rows; the result is tested on your files, reviewed by someone other than its author, versioned, then run without a model. A change is replayed on every past period kept before it is approved. When a file changes format, the run stops and a repair is proposed, for approval. A sealed audit pack traces versions and approvals.
What happens when a file's format changes?
Before each run, Nestor compares the file's structure with the reference one. If it changed, the run stops and explains the difference. For single-step agents written by Nestor, a repair can be proposed; someone approves it before it goes back into service.
Is Nestor shared with other companies?
No. Nestor is not a shared service: it is installed on an instance dedicated to your company (in Europe, or on your infrastructure), with its own data and secrets. Inside the company, teams are isolated from one another.
How do I get started?
Request an audit: we start from your real tasks to scope AI in your company. You can also book a demo of Nestor.
The beaver builds the dam. Your company reaps the rewards.
Request an audit of how your company uses AI (tokens, products, AI Act), or book a demo of Nestor.