Our certifications
Security is not a statement: it is a management system audited every year by an independent third party.
Every model runs on our own infrastructure. No prompt, document or piece of your company data is ever sent to OpenAI, Anthropic, Google or Microsoft.
Many companies say private AI when they mean we use the OpenAI API under a business contract: your documents still leave the building and get processed on a non-European provider servers. That is not the same thing. The question to ask is not whether the system is private, but where the model runs and what leaves your perimeter. Below is our answer, in writing.
Can you guarantee that all LLMs, embeddings, OCR, transcription, RAG and document processing run exclusively on servers under our control, without sending any data, prompt, document, embedding or metadata to OpenAI, Anthropic, Google, Microsoft or any other external AI provider? Can the system work completely without an internet connection?
Yes. The entire processing chain runs on hardware controlled by us or by you: language models, embeddings, vector database, OCR, audio transcription and computer vision. There is no call to any general-purpose AI API at any point in the flow. In the on-premise setup the system also works with the internet unplugged, and we demonstrate it during acceptance testing. We put these commitments in the contract, not just on a web page.
Every component of the chain runs in-house. None of these steps calls an external service.
Open-weight models running on our own GPUs. We choose and update the model together with you, and the weights stay on the project infrastructure.
Your documents are vectorised locally. Embeddings are as sensitive as the documents they come from: they never leave the perimeter.
Vector database on our own hardware. The assistant answers citing your documents, with nothing indexed anywhere else.
PDFs, scans, invoices, contracts and images are read by OCR engines installed on the server. No document is ever uploaded to a third-party cloud service.
Meetings, calls and voice notes transcribed locally. The synthetic voice of our callbots is generated on our infrastructure too.
Image and video recognition and cataloguing on our servers, with no external vision APIs.
Sales promises protect nobody. These commitments go into the contract, before the project starts.
We sign the confidentiality agreement before you show us a single file. It covers documents, data, processes, prices and everything we see along the way, and it stays valid even if we end up not working together. If you prefer your own NDA template, we sign yours.
On request we configure the system to keep nothing: prompts, documents and answers are processed in memory and discarded at the end of the session, with no persistent logs and no conversation history. It is a verifiable configuration, and we write it into the contract with the related guarantees.
The fact that no prompt, document, embedding or metadata is sent to external AI providers is not a marketing statement: it is a contract clause, listing the components in use and where they run. Your legal counsel or DPO can review it before signing.
The server sits at your premises, inside your network. It can run completely offline, with no internet connection: this is the setup for healthcare, industrial or confidential data.
Hardware reserved for your project in a European datacenter, managed by us. No resource sharing with other clients.
An isolated environment on European infrastructure, to get started quickly at a contained cost. Data stays within the European Union.
Security is not a statement: it is a management system audited every year by an independent third party.
Ask us the hard questions: which models we use, where they run, who has access, what happens to the data at the end of the project. We answer in writing, before you show us anything.
ISO 27001:2022Information Security
ISO 27017:2015Cloud Security
ISO 27018:2019Privacy in Cloud
ISO 9001:2015Quality Management