As little as possible, well protected
Read-only access by default, dedicated accounts, secrets kept out of the code, strong authentication on every tool we deliver.
Models
- Claude, Gemini, GPT
- Open models, run locally
- Chosen on cost and quality
- Cost measured per task
Agents
- Tools and connectors (MCP)
- Multi-agent orchestration
- Human approval
- Every action logged
Data
- ERP, CRM, SQL databases
- APIs, feeds, files
- Read-only by default
- Quality, traceability
Cloud
- AWS, Azure, Google Cloud
- Continuous deployment
- Protected access and secrets
- FinOps and efficiency
Where to run AI
Three ways to host a model, three trade-offs.
The choice depends on your data more than on fashion. We make it with you, and we can combine all three in a single project.
| Option | Quality | Cost | Confidentiality | When to choose it |
|---|---|---|---|---|
| Leading model via API | Pay as you go, low to start | Complex reasoning, writing, analysis. Non-sensitive data, or a suitable contract. | ||
| Private cloud or dedicated region | Higher, predictable | Regulated data, hosting in Europe required, large volumes. | ||
| Local model | Hardware, then close to zero | Data that must stay in-house, offline use, focused and repetitive tasks. |
Security and compliance
Built in from day one, not at the end.
No unnecessary personal data
We only extract what is needed. Personal data is excluded or pseudonymised, and nothing is reused to train a model.
GDPR and provider terms
A record of processing activities, appropriate hosting, and a review of each AI service's terms of use before we use it.