Skip to main content
Knowledge bases let your scenario retrieve facts from your documents during a live conversation. Use them when your assistant should answer with domain-specific information instead of relying only on general model knowledge.
Throughout the Akapulu platform and docs, the terms corpus and knowledge base are used interchangeably.

​
Overview of knowledge bases

A knowledge base contains:
  • a knowledge base (container)
  • one or more documents (your uploaded files)
During a conversation, your scenario can call a RAG tool function that queries the selected knowledge base and returns relevant context to the model before it responds.

​
Create a knowledge base

  1. Go to akapulu.com/knowledge-bases.
  2. Click New Knowledge Base.
  3. Enter:
    • name
    • optional description
  4. Open the new knowledge base and click Add Document +.
  5. set:
    • document name
    • optional document description
  6. Upload a file
Save and wait for processing to complete. After upload, Akapulu creates searchable vector chunks from the document so the knowledge base can be used in RAG calls.

​
Attach a knowledge base to a scenario

​
Attach via JSON

Attach the knowledge base in your scenario by adding a function with type: "rag" and the target corpus_id: The function name and description are presented to the LLM as tool metadata, so use clear action-oriented wording. 
{
  "function": {
    "name": "<knowledge_base_name>",
    "description": "<description>",
    "type": "rag",
    "corpus_id": "<KNOWLEDGE_BASE_ID>"
  }
}
For example: 
{
  "function": {
    "name": "lookup_insurance_policy",
    "description": "Search the insurance policy knowledge base for coverage details and return relevant excerpts.",
    "type": "rag",
    "corpus_id": "9a7d5d72-4f31-4b9b-8b7a-f3091da2ab6d"
  }
}

​
Attach via UI

Use the scenario builder when you want to attach a knowledge base without editing JSON:
  1. Open your scenario and go to the target node.
  2. Click Add function.
  3. In the modal, open the RAG Tool tab.
  4. Select your knowledge base by name.
  5. Enter a semantically meaningful function name and description.
The function name and description are shown to the LLM as tool metadata, so they should clearly describe when the tool should be called. Attach a knowledge base as a RAG tool (1.5x speed) Shown at 1.5x speed. When this function is included in a node:
  • the model can invoke it like any other tool
  • Akapulu queries the specified knowledge base
  • relevant context is returned to the LLM to use in its response