What you get
A controlled context layer that connects agents to your company's wikis, documents, repositories, project decisions, and operational systems with citations, freshness controls, and role-based retrieval. The model still does the thinking; Company Memory does the remembering.
- Canonical source registry with per-document permissions.
- Permission-aware retrieval with citations on every answer.
- Freshness signals so stale knowledge cannot silently win.
- Working memory and session memory, scoped per agent and per user.
- Evaluation hooks for retrieval quality and citation accuracy.
How it fits the rest of Attnora
Company Memory is the source of truth that the agent layer reads from and the AI gateway caches and routes around. It is the same control surface that governance evaluates against.
How a pilot starts
We map one real workflow with three to five authoritative sources, define roles and citation expectations, and ship a working memory in two to four weeks. Expansion follows evidence.
Frequently asked questions
Is Company Memory a vector database?
No. It is the control layer that owns sources, permissions, citations, and freshness, and it uses vector and keyword retrieval underneath. The vendor choice for those primitives is open.
Can we keep our existing wiki and document store?
Yes. Company Memory reads from your canonical systems rather than replacing them. Source priority, citations, and freshness are tracked per source.
How is access controlled?
Permissions are inherited from the underlying source. An agent cannot retrieve a document its user would not be allowed to read in the source system.
What languages and formats are supported?
German and English are first-class. Confluence, Notion, SharePoint, GitHub, PDF, and Office formats are supported through the same connector layer.
Related services
Next step
Map one workflow into Company Memory.
Tell us which sources and which user roles matter first. We will respond within one business day.
Request a discovery session