Agent Knowledge Base

A knowledge base lets your agent answer questions from your actual documents — product manuals, internal wikis, FAQs, legal policies — rather than relying solely on the AI's training data.

Attaching documents

Documents must already be uploaded to your Documents library before you can attach them to an agent.

1. Open the agent builder and click the Knowledge tab.
2. Click Add documents.
3. Browse or search your document library.
4. Select up to 20 documents and click Attach.

The documents appear in a list on the Knowledge tab with their indexing status (Indexed, Pending, or Failed).

How the agent uses documents

When a user sends a message to the agent, the RAG pipeline runs automatically:

1. The user's message is converted into a vector embedding.
2. Hubrix searches the agent's knowledge base for the most semantically similar document chunks.
3. The top chunks are injected into the AI's context alongside the message.
4. The AI answers using both its training knowledge and your document content.

Maximum documents

Each agent can have up to 20 documents attached. If you need more, consider splitting the use case across two specialised agents (for example, one for product docs and one for legal policies).

The Reindex button

If a document's content changes — for example, you re-uploaded a newer version — click Reindex next to that document on the Knowledge tab. This re-processes the document and rebuilds its vector embeddings so the agent uses the latest version.

You may also need to reindex if a document shows a Failed status after the initial attachment.

Tips for a useful knowledge base

• Upload clean text documents — text-based PDFs and DOCX files extract far better than scanned images.
• Name your documents clearly — the file name is visible in citation references, so acme-return-policy-2026.pdf is more informative than scan001.pdf.
• Keep documents focused — a document that covers one topic retrieves more accurately than a sprawling 200-page all-in-one manual.
• Remove stale documents — detach documents that are no longer accurate. Outdated information retrieved by RAG is worse than no information.