This feature allows your agent to retrieve contextual information from a knowledge base (called a Library) during live conversations. It’s typically used to:

1. Answer complex or policy-driven queries by referencing internal documents.

2. Reduce hallucinations by grounding responses in verified company knowledge.

This enables your agent to handle a wider range of queries confidently, without relying solely on training data or pre-scripted responses.

Create and manage library

1. Navigate → Library → Create Library from the main menu.

2. Click + Add Document and:

   • Upload a PDF (e.g., SOP, Policy Doc, Product Manual), or

   • Create a new document using the built-in editor.

3. Name your document clearly (e.g. Returns Policy - March 2024).

4. To update a document:

   • Upload a new version.

   • Delete the outdated one.

A single Library can contain multiple documents.

Enable in agent

1. Navigate → Agents → Select Agent → Edit → Think Settings.

2. Scroll to Library Access section and toggle enabled.

3. Select:

   • Tool Name: this identifies the tool inside the agent (e.g. retrieve_context).

   • Library: choose the library you created.

   • Prompt: explain when the agent should access the library (e.g., "If user asks for refund policy").

   • Similarity Threshold: adjust relevance sensitivity (default: 0.75).

   • Number of Documents: max results returned for each search (default: 3).

   • Messages (Optional): define what the agent says when searching or interrupted.

How it works during call

• When the agent receives a user query during the call, it checks your Library Prompt for matching instructions.

• If triggered, it will retrieve the most relevant documents based on your configured similarity threshold.

• The agent then incorporates that document content into its reply.

Tips and best practices

• Document formatting matters: ensure PDFs or text docs are clear, structured, and easy to chunk semantically.

• One topic per doc: split large files into smaller topic-specific documents.

• Use Prompt smartly: clearly define when the library should be used (e.g., refund, warranty, escalation).

• Limit scope: avoid uploading large, general-purpose documents.

• Test with sample queries to verify relevance and responses.

Example

Library Prompt:

If the user asks a question related to return, exchange, refund, or cancellation policies, search the library and respond using the content retrieved.

Similarity Threshold: 0.75

Max Documents: 3

Message (optional):

• Before search: "Let me check that for you."

• On interruption: "Just a second, finishing up my search."