MongoDB-native AI agent

Demo AI agent using MongoDB, Voyage AI, and OpenAI that implements agentic RAG and agent memory. No agent-specific frameworks needed.

Tools: vector_seach_tool and calculator_tool

About

Workflow:

• Chunk and ingest vector embeddings into MongoDB Atlas. Uses LangChain for loading/chunking.
• Specify the session id (agent will persist conversation interactions in a MongoDB collection).
• Run queries about sample data (recent MongoDB earnings report) or math operations to use tools.

Models used:

• Voyage AI: voyage-3-large
• OpenAI: gpt-4o

Prompt engineering methods used:

• reAct + few-shot prompting

Configuration

1. Install dependencies:

pip install --quiet --upgrade pymongo voyageai openai langchain langchain_mongodb langchain_community

2. Create .env file with necessary credentials (MongoDB Atlas connection string, Voyage AI, and OpenAI API keys)

3. Run the agent:

python main.py