Knowledge & Data
Knowledge Bases

Knowledge Bases

Knowledge Bases hold the content your agent uses to answer questions — website pages, PDFs, help articles, policies, FAQs. The AI searches this content using semantic search (RAG) to find relevant answers.

Creating a knowledge base

Go to Knowledge & Data in the left nav → New Knowledge Base. Give it a name.

Adding sources

Click Add Source inside any knowledge base. Choose a type:

Website — enter a URL and choose crawl depth: this page only, this page + subpages, or full site.

PDF Document — upload one or more PDFs. Text is extracted and indexed automatically.

CSV / Excel — upload a spreadsheet. Good for reference data like pricing sheets. For product inventory, use a Product Catalog instead.

Text / Markdown — paste or type content directly. Good for policies, FAQs, or any content that doesn't exist as a file.

Q&A Pairs — add structured question-and-answer pairs. The most reliable format for FAQ content — the AI matches user questions directly to your answers.

Google Drive — connect your Google Drive to import Docs, Sheets, PDFs, and text files directly. Go to Connections to set up the integration first, then use the Drive file picker inside any KB.

Instagram / Facebook — connect your Meta accounts to sync your social media content (bio, posts) as knowledge sources automatically.

Notion — connect Notion to import pages and databases. You can select specific pages to sync or import everything.

How search works

When a user asks your agent a question, Zimmer uses semantic search (RAG) to find the most relevant content:

  1. The user's question is converted to a vector embedding
  2. pgvector finds the most similar chunks from your knowledge bases
  3. The top matching chunks are injected into the AI's context
  4. The AI generates an answer grounded in your actual content

This means your agent answers based on your real documentation — not generic AI knowledge.

Training

After adding or updating sources, click Update Training. Status shows:

  • Trained — ready
  • Training — indexing in progress (usually under a minute)
  • Error — check the source URL or file

Linking to agents

Agent → Knowledge tab → select the KBs this agent should use. An agent can use multiple knowledge bases.

Tips

  • Keep each KB focused on one topic — easier to update and more accurate retrieval
  • Q&A pairs work best where exact question wording matters
  • Re-train after any significant content update