This guide explains how to load large context into the Active Knowledge Base (AKB) by splitting it into focused topics the AI can search and use. Following this process improves retrieval quality and ensures the right context is injected into prompts for accurate answers.
When to use AKB topics
Use AKB topics whenever a single attribute’s context would exceed roughly two/three pages of text.
Tip: Smaller, well-named topics are easier for the AI to select correctly based on user intent.
Prepare the source document
Split the customer’s document (e.g., 100 pages) into multiple, narrowly scoped sections.
Make each section about one specific topic/concept.
Give each section a clear, specific name that describes exactly what it covers.
Prefer English where possible; other languages are supported but English is recommended for best LLM performance.
For tables, use Markdown language.
Example naming:
Let's say you have a section of text describing the security protocols of a business as it relates to company-owned laptops. Instead of naming the section just "Security protocols," be more specific, such as "Security protocols for company-owned laptops."
Navigate to the AKB
From the Organizations page, click the three-dot icon on your customer's AI Employee and Go to Builder.
Navigate to the AKB page.
In the top dropdown, select GeneralManagerAgent and click Save.
Create an AKB topic
Click Add New Topic.
Fill in the fields:
Name: Enter the descriptive name of the first section.
Facts: Enter the same value as Name.
Summary: Paste the full text of that section.
Source: Provide a unique incremental value (e.g.,
R1,text1,topic_1).Labels: Enter
rag_context(case sensitive; all lowercase).
Click Create.
Add the remaining sections
Repeat Create an AKB topic for every section you split from the original document.
Increment the Source value each time (e.g.,
R2,R3, … ortopic_2,topic_3, …).Ensure each topic covers only one idea and has a unique, precise Name.
How retrieval works
When a user asks a question, the AI evaluates the Name of AKB topics to find the closest match to the user’s intent. It then injects that topic’s Summary into the prompt to answer the question.
Future roadmap
An automated feature to split long documents and create AKB topics is planned. Until then, follow the manual process above.