Knowledge Base

Give agents your own documents to answer from, retrieved on demand with RAG.

The knowledge base holds the files, web pages, and text your agents can draw on. At run time an agent retrieves the most relevant pieces (RAG) instead of relying on the model's general training.

Add knowledge

  1. In the sidebar, open Content & memory → Knowledge Base.
  2. Add a source with the toolbar:
    • Add Files — upload PDF, DOCX, TXT, HTML, Markdown, or EPUB (up to 21 MB each).
    • Add URL — index a web page by its address.
    • Create Text — paste content directly with a Name and Content.
    • Create Folder — group related items.
  3. Watch the Status column move from Processing to Completed — the item is chunked and embedded, then ready to retrieve.

Manage and tune

Search or filter by Type and Status. Each row's menu offers Download, Reprocess (regenerate embeddings), and Delete. Click Configure RAG to set the embedding model and retrieval limits, and Playground to test a query and see which chunks come back.

Let an agent use it

Open an agent version and attach the knowledge sources it should retrieve from; the agent then answers using that content.

Tip: If answers ignore a document, check its Status is Completed and that the source is attached to the agent — an item that only lives in the knowledge base is never retrieved on its own.