Agentic RAG Agent

Agentic RAG Agent is a chat application that combines models with retrieval-augmented generation. It allows users to ask questions based on custom knowledge bases, documents, and web data, retrieve context-aware answers, and maintain chat history across sessions.

Note: Fork and clone this repository if needed

1. Create a virtual environment

uv venv .venv
.venv\Scripts\activate

2. Install dependencies

uv pip install -r requirements.txt

3. Configure API Keys

Required:

export GROQ_API_KEY=your_groq_key_here

Optional (for additional models):

export ANTHROPIC_API_KEY=your_anthropic_key_here
export GOOGLE_API_KEY=your_google_key_here

For Hugging Face API embeddings (recommended to save disk space):

export HUGGINGFACE_API_KEY=your_huggingface_key_here

Note: If HUGGINGFACE_API_KEY is set, the application will use the Hugging Face API for embeddings instead of downloading the local BGE model (saves ~1.3GB of disk space). If not set, it will fall back to the local BGE embedder.

4. Run PgVector

Install docker desktop first.

• Run using a helper script (make it executable first)

chmod +x run_pgvector.sh
./run_pgvector.sh

• If you get a container name conflict error, you can either:

  • Use the existing container if it's the same image: docker start pgvector
  • Remove the old container: docker rm pgvector and then run the script again

• OR run using the docker run command

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

5. Run Agentic RAG App

streamlit run app.py

🔧 Customization

Model Selection

LLM Providers

The application supports multiple LLM providers:

• Groq (meta-llama/llama-4-scout-17b-16e-instruct, llama-3.3-70b-versatile) - Default
• OpenAI (o3-mini, gpt-4o)
• Anthropic (claude-3-5-sonnet)
• Google (gemini-2.0-flash-exp)

Embedding Model

The application supports two embedding options:

1. Hugging Face API (Recommended)

   • Model: BAAI/bge-large-en-v1.5
   • Dimensions: 1024
   • Benefits: No local model download required (saves ~1.3GB disk space), high-quality embeddings
   • Requirements: Hugging Face API key (set as HUGGINGFACE_API_KEY environment variable)

2. Local BGE Model (Fallback)

   • Model: BAAI/bge-large-en-v1.5
   • Dimensions: 1024
   • Benefits: No API key required, works offline
   • Drawbacks: Requires ~1.3GB disk space for model download

How to Use

• Open localhost:8501 in your browser.
• Upload documents or provide URLs (websites, csv, txt, and PDFs) to build a knowledge base.
• Enter questions in the chat interface and get context-aware answers.
• The app can also answer question using duckduckgo search without any external documents added.

Troubleshooting

• Docker Connection Refused: Ensure pgvector container is running (docker ps).
• Container Name Conflict: If you get an error about the container name already in use, see the instructions in the "Run PgVector" section.
• Groq API Errors: Verify that the GROQ_API_KEY is set and valid.
• Hugging Face API Errors: If you see errors related to the Hugging Face API, check that your HUGGINGFACE_API_KEY is valid and that you have access to the model. You may need to accept the model's terms of use on the Hugging Face website.
• Embedding Model Issues:
  • API Mode: If using Hugging Face API and encountering errors, check your API key and internet connection.
  • Local Mode: If using the local BGE embedder, ensure you have enough disk space for the model download (~1.3GB) and that your Python environment has the required dependencies installed.

📚 Documentation

For more detailed information:

• Agno Documentation
• Streamlit Documentation