Knowledge base

Ground Clara’s answers in your pitch decks, product docs, FAQs, and website with retrieval-augmented generation.

A knowledge base is a collection of your content - documents, plain text, and website pages - that Clara searches when answering buyer questions. Instead of generic model knowledge, Clara answers from your material, in your voice.

Knowledge bases are reusable: attach one knowledge base to many agents, or give each agent its own.

Content types

Type	What it is	Best for
Documents	PDF and DOCX files	Pitch decks, product docs, policies, reports
Plain text	Pasted raw text	FAQs, scripted answers, word-for-word control
URL & website	Single pages or a full site crawl of your docs	Public docs, help centers, product pages

How retrieval works

Two retrieval modes are available:

• Agentic retrieval (recommended) - Clara decides when and which knowledge base to search based on the question. More precise, less noise.
• Traditional retrieval - every query searches all attached knowledge bases. Simpler and more predictable.

With agentic retrieval, the knowledge base name and description matter: Clara reads them to decide whether to search it. Name it for the domain - Product-Docs, Pricing-2026 - and describe when to use it.

A vague name or empty description means Clara may never pick the knowledge base, no matter how good the content is.

Creating a knowledge base

1. In your workspace, open Knowledge base.
2. Click Create New.
3. Add a name and a description that says when Clara should use it - for example: “Use this whenever the buyer asks about pricing, plans, or limits.”
4. Add content: upload files, paste text, or add URLs.
5. Save & Create. Content is queued for processing: extraction, cleaning, chunking, and indexing run automatically.

Organizing knowledge bases

Guideline	Why
One knowledge base per domain	Focused collections retrieve more accurately
Around 5–50 documents each	Quality drops sharply once mixed content exceeds ~50 documents
Split by topic, not by team	Clara picks the right base by name and description
Descriptive file names	File names appear in retrieval reasoning - pricing-2026.pdf beats document-final-v2.pdf

Ten focused documents per knowledge base beat one giant mixed library. Multiple small knowledge bases beat one catch-all.

Attaching to agents

1. Open the agent and go to the Knowledge base section.
2. Select one or more knowledge bases.
3. Save. Changes apply on the next conversation.

Multiple agents can share the same knowledge base - update the content once and every attached agent picks it up on the next session.

Content best practices

• Prefer clean text over complex PDFs - multi-column and footnote-heavy layouts retrieve poorly
• Split large material into smaller documents; they retrieve better
• Re-upload changed documents rather than editing in place
• Large files and full-site crawls can take a few minutes to index; content answers correctly once indexed