Knowledge Bases
A knowledge base provides an AI agent with reference documents — catalogs, terms and conditions, price lists, template contracts — so its checks take your organization's business context into account rather than relying solely on the model's general knowledge (RAG — Retrieval-Augmented Generation).
Creating and populating a base
From the Manager → Knowledge Bases tab:
- Create a base (name, embedding model used for vectorization, segment size and overlap — 1000 characters and 200 overlap by default).
- Upload reference documents — accepted formats:
.docx,.pdf,.pptx,.txt,.md. - Each document is automatically split and vectorized in the background; the base shows its last indexing date.
Documents are physically stored in a RAG storage space attached to the base (see Managing RAG Storage Spaces) — either platform-managed storage, or an S3/SharePoint space brought by your organization.
Link with AI agents
An AI agent is attached to a knowledge base in its configuration. During analysis, each segment of the document being checked automatically queries the base to retrieve the most relevant reference passages, which are injected into the context sent to the AI.
An agent attached to a knowledge base takes precedence over any document index node present in the pipeline — the two mechanisms can coexist but serve different needs: the knowledge base brings shared, external business context, while the document index node lets you retrieve related passages within the document being analyzed itself.
Access rights
Uploading and browsing documents is open to any management profile; deleting a base or its documents is restricted to the kb_manager role (or a legacy role).