Agentic RAG

This example application shows how to build a sophisticated RAG (Retrieval Augmented Generation) system that leverages search of a knowledge base with LLMs to provide deep insights into the data.

​

The agent can:

• Process and understand documents from multiple sources (PDFs, websites, text files)
• Build a searchable knowledge base using vector embeddings
• Maintain conversation context and memory across sessions
• Provide relevant citations and sources for its responses
• Generate summaries and extract key insights
• Answer follow-up questions and clarifications

​

The agent uses:

• Vector similarity search for relevant document retrieval
• Conversation memory for contextual responses
• Citation tracking for source attribution
• Dynamic knowledge base updates

​

Example queries to try:

• “What are the key points from this document?”
• “Can you summarize the main arguments and supporting evidence?”
• “What are the important statistics and findings?”
• “How does this relate to [topic X]?”
• “What are the limitations or gaps in this analysis?”
• “Can you explain [concept X] in more detail?”
• “What other sources support or contradict these claims?”

​

Code

The complete code is available in the Agno repository.

​

Usage

git clone https://github.com/agno-agi/agno.git
cd agno

python3 -m venv .venv
source .venv/bin/activate

pip install -r cookbook/examples/apps/agentic_rag/requirements.txt

./cookbook/scripts/run_pgvector.sh

docker run -d \
  -e POSTGRES_DB=ai \
  -e POSTGRES_USER=ai \
  -e POSTGRES_PASSWORD=ai \
  -e PGDATA=/var/lib/postgresql/data/pgdata \
  -v pgvolume:/var/lib/postgresql/data \
  -p 5532:5432 \
  --name pgvector \
  agnohq/pgvector:16

# Required
export OPENAI_API_KEY=***
# Optional
export ANTHROPIC_API_KEY=***
export GOOGLE_API_KEY=***

streamlit run cookbook/examples/apps/agentic_rag/app.py

Need help? Join our Discourse community for support!