Your company knowledge. In the right working context.
Instructions, product information and operational experience matter when they reach the right task. Brezel connects approved knowledge with responsibilities and access permissions.
01 /
What belongs in an AI knowledge base?
Useful material includes information your team needs repeatedly: approved instructions, product documents, templates and business guidelines. Relevant records and case histories can also provide working context.
Before adding content, establish its owner, currency and permitted audience. More documents alone do not automatically improve answers.
Conceptual diagram
01Company guidelines
02Approved team knowledge
03Personal context
This diagram explains the relationship. It is not a live view or a commitment to future features.
02 /
Company, department and personal knowledge.
Management guidelines establish the company-wide framework. Departments add their ways of working. Personal and case-specific information stays within its intended scope. Knowledge is not a single collection open to everyone.
The source material distinguishes Company Info, Playbooks and Memory: set business guidelines, reviewed ways of working, and relevant facts or prior context. The complete automatic learning loop and role inheritance are still being developed.
03 /
Share approved knowledge within clear boundaries.
Generally approved guidance can be passed down to relevant roles. This does not automatically send personal or departmental information back to every higher level. Cross-department use requires a defined purpose and permitted scope.
A case observation does not become company policy merely because it was stored. Relevance, confidentiality and ownership need review before wider adoption. Particularly sensitive content requires separate treatment.
04 /
Retrieve knowledge rather than train the base model.
Relevant approved material is supplied for each task. Company data is not used to train the underlying AI model. The model receives necessary context within the configured processing arrangement.
Cloud and today’s on-premise offering use an external European language model service. A local Brezel system therefore does not yet mean all information needed for AI tasks is processed locally.
05 /
Knowledge needs maintenance.
Assign owners, keep important documents current and review access after role changes. AI can reuse incorrect or outdated guidance. Results need checking appropriate to their use.
Answer-level source references, continuous handling of changed documents and permissions, and automatic distillation of experience are development goals. Today’s package scope is agreed separately.