awesome-llm-apps/rag_tutorials/contextualai_rag_agent/README.md

Contextual AI RAG Agent

A Streamlit app that integrates Contextual AI's managed RAG platform. Create a datastore, ingest documents, spin up an agent, and chat grounded on your data.

Features

• Document ingestion to Contextual AI datastores
• Agent creation bound to one or more datastores
• Response generation via Contextual’s Grounded Language Model (GLM) for faithful, retrieval-grounded answers
• Reranking of retrieved documents by query relevance and custom instructions (multilingual)
• Retrieval visualization (show attribution page image and metadata)
• LMUnit evaluation of answers using a custom rubric

Prerequisites

• Contextual AI account and API key (Dashboard → API Keys)

Generate an API key

1. Log in to your tenant at app.contextual.ai.
2. Click on "API Keys".
3. Click on "Create API Key".
4. Copy the key and paste it into the app sidebar when prompted.

How to Run

1. Clone the repository and navigate to the app folder:

git clone https://github.com/Shubhamsaboo/awesome-llm-apps.git
cd awesome-llm-apps/rag_tutorials/contextualai_rag_agent

2. Create and activate a virtual environment.
3. Install dependencies:

pip install -r requirements.txt

4. Launch the app:

streamlit run contextualai_rag_agent.py

Usage

1. In the sidebar, paste your Contextual AI API key. Optionally provide an existing Agent ID and/or Datastore ID if you already have them.

2. If needed, create a new datastore. Upload PDFs or text files to ingest. The app waits until documents finish processing.

3. Create a new agent (or use an existing one) linked to the datastore.

4. Ask questions in the chat input. Responses are generated by your Contextual AI agent.

5. Optional advanced features:

   • Agent Settings: Update the agent system prompt via the UI.
   • Debug & Evaluation: Toggle retrieval info to view attributions; run LMUnit evaluation on the last answer with a custom rubric.

Configuration Notes

• If you're on a non-US cloud instance, set the Base URL in the sidebar (e.g., http://api.contextual.ai/v1). The app will use this base URL for all API calls, including readiness polling.
• Retrieval visualization uses agents.query.retrieval_info to fetch base64 page images and displays them directly.
• LMUnit evaluation uses lmunit.create to score the last answer against your rubric.