Talking to your Data: The RAG APIs

Once your Knowledge Base (KB) is synced, you can query it using two different methods: retrieve (Just the facts) and retrieve_and_generate (The answer).

1. retrieve (The Raw Facts)

Use this if you want to handle the "Generation" yourself (e.g., using a custom prompt).

• Output: A list of the most relevant chunks from your PDFs.
• Why?: Total control over how the AI uses the data.

2. retrieve_and_generate (The Full Loop)

This is a "One-Stop-Shop" API. It finds the facts, tells the AI to read them, and gives you a complete answer in one call.

import boto3

client = boto3.client("bedrock-agent-runtime", region_name="us-east-1")

response = client.retrieve_and_generate(
    input={"text": "What is the holiday policy?"},
    retrieveAndGenerateConfiguration={
        "type": "KNOWLEDGE_BASE",
        "knowledgeBaseConfiguration": {
            "knowledgeBaseId": "YOUR_KB_ID",
            "modelArn": "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-haiku-20240307-v1:0"
        }
    }
)

print(response["output"]["text"])

---

3. Visualizing the Decision

graph TD
    User[Question] --> Mode{Choose Mode}
    Mode -->|Retrieve Only| R[Raw Chunks]
    R --> Custom[Custom Logic/Tool]
    
    Mode -->|Retrieve & Generate| RG[Managed Brain]
    RG --> Final[Final Human Answer]

---

4. Why retrieve_and_generate is usually better

• It automatically handles the System Prompt for RAG.
• It is optimized for Grounding (preventing the AI from knowing things outside your PDFs).
• It handles the Citations (Module 8 Lesson 2) automatically.

---

Summary

• retrieve returns raw text segments.
• retrieve_and_generate returns a final answer and citations.
• Both APIs require the bedrock-agent-runtime client.
• Managed RAG is faster to build and easier to maintain.