Praison AI

Get Started

RAG & Knowledge

Add knowledge bases and retrieval-augmented generation to your agents

Next

Learn how to enhance your agents with knowledge bases and retrieval-augmented generation (RAG).

python
from praisonaiagents import Agent, KnowledgeConfig

agent = Agent(
    name="Knowledge Assistant",
    instructions="Answer from the knowledge base.",
    knowledge=KnowledgeConfig(sources=["./docs"]),
)

agent.start("Find the section about session persistence.")

The user attaches a knowledge source, asks a question, and the agent grounds its reply in retrieved passages.

Diagram

RAG

📋 Question

🔍 Retrieve

✅ Grounded Answer

Quick Start

  1. Simple Usage

    Point the agent at a folder of documents and it answers from them.

    python
    from praisonaiagents import Agent, KnowledgeConfig

agent = Agent(
    name="Knowledge Assistant",
    instructions="Answer from the knowledge base.",
    knowledge=KnowledgeConfig(sources=["./docs"]),
)
agent.start("Find the section about session persistence.")
  2. With Configuration

    Tune how many passages come back and how similar they must be.

    python
    from praisonaiagents import Agent, KnowledgeConfig

agent = Agent(
    name="Knowledge Assistant",
    instructions="Cite retrieved passages.",
    knowledge=KnowledgeConfig(sources=["./docs"], top_k=5, similarity_threshold=0.7),
)
agent.start("Summarise the security section.")

How It Works

Diagram
KnowledgeAgentUserKnowledgeAgentUserAsk a questionRetrieve top-k passagesMatching chunksAnswer grounded in sources

Related

Knowledge Base Setup

Create and configure knowledge bases

 Chunking Strategies

Optimize document chunking

 Retrieval Methods

Configure retrieval strategies

Best Practices

Start with a folder of clean sources

KnowledgeConfig(sources=["./docs"]) indexes a directory directly. Remove boilerplate and duplicates first so retrieval returns signal, not noise.

Raise the similarity threshold to cut noise

A higher similarity_threshold filters weak matches. Start at 0.7 and adjust based on whether answers miss context or cite irrelevant passages.

Keep top_k small for focused answers

Fewer, stronger passages usually beat many weak ones. Increase top_k only when answers lack coverage.