Sovereignty
Models run within your infrastructure and can operate without an internet connection. No data is sent to an external API.
Evidence to require
The delivered architecture, and a demonstration with the network disconnected.
Commitments
Six commitments, each one verifiable.
Sovereignty, measured reliability, reversibility, traceability, fit to your context, a single point of contact: these commitments are not statements of intent. Each comes with the evidence you are entitled to require: where the models run, the evaluation report, the handover of the code, the source of every answer. What enters your perimeter stays there, and you can check it.
Measured reliability
Every product is delivered with an evaluation report: answer accuracy, correctness of refusals when the information does not exist, quality of citations.
Evidence to require
The evaluation report, with its question set and measurement context.
Reversibility
You receive the code, documented installation and a skills transfer. Your data, configurations and content remain under your control; the architecture and documentation are designed to avoid any irreversible dependency on Oppdrag.tech. You can evolve the product without us.
Evidence to require
The handed-over code repository, and the installation documentation.
Traceability
When an answer relies on documents, the sources cited can be found and checked: document, page, passage. A colleague can always check what the product states.
Evidence to require
Any answer, picked at random, and the passage it cites.
Fit to your context
Products are configured to your data, your processes, your users and your operational constraints. A project starts with scoping, and each stage is decided on results.
Evidence to require
The scoping note, which sets the perimeter and schedule before any commitment.
A single point of contact
The person who scopes your need also understands the architecture and the technical choices of the product delivered. When the project requires it, verified experts are brought in, selected to the same standard.
Evidence to require
The same point of contact, from scoping to handover, who masters the architecture of what they present to you.
Recognised frameworks
Knowledge AI is designed according to the requirements of ISO/IEC 42001 on artificial intelligence management systems, and follows OWASP recommendations for application security. It is not certified; its design is aligned with these requirements.
What we do not claim
We announce neither absolute success rates nor guaranteed lead times while our project history cannot support them. Lead times are set out at scoping, and performance is measured on your data.
Check for yourself
Each commitment can be verified during a Knowledge AI demonstration, on documents you choose.