Knowledge base

A knowledge base is the source material the agent draws on to answer questions it could not answer from the prompt alone — your pricing, your policies, your FAQ. You add sources, the platform splits and indexes them, and the agent retrieves the relevant pieces when a caller asks.

Adding sources

There are three ways to add a source:

• Upload — a file. PDF, Word, plain text, CSV, or markdown.
• URL — pull the content from a web page.
• Text — paste a snippet or article directly.

Each source is broken into chunks and embedded so it can be searched by meaning, not just keywords. You do not manage the chunks — adding the source is all you do.

How the agent uses it

The knowledge base is wired to the agent through the Search knowledge base builtin tool. When a caller asks something the prompt does not cover, the model calls that tool, gets back the most relevant chunks, and uses them to ground its answer. So:

• The knowledge base only helps if the Search knowledge base tool is on the agent.
• The tool's description should tell the model when to search — "use this when the caller asks about plans, pricing, or policy".

What to put in it

Put the things callers actually ask about and that change often enough that baking them into the prompt would be a maintenance headache:

• Pricing and plan details.
• Policies — refunds, cancellations, eligibility.
• Product or service specifics.
• A FAQ.

Keep the prompt for persona and rules; keep facts in the knowledge base. That way you can update a price by editing one source instead of rewriting the prompt.

Keeping it fresh

When a source changes, update it here. The agent answers from what is indexed, so stale sources mean stale answers — treat the knowledge base as living content, not a one-time upload.