Build AI agents for real-time financial options analysis with LangGraph, ChromaDB RAG, and Polygon.io data
This skill describes building an AI agent for real-time financial options analysis using LangGraph, ChromaDB, and Polygon.io market data. It addresses the problem of assembling an intelligent trading-research assistant that can retrieve options data, cache it efficiently, retain conversation context across sessions, and produce professional-grade analysis. The project is delivered as a complete agent system cloned/configured from an upstream layout, combining real-time data retrieval with retrieval-augmented generation and persistent memory.
Its capabilities include real-time options data retrieval from Polygon.io with intelligent caching, a RAG knowledge base backed by ChromaDB for semantic search over cached data, persistent conversation memory via SQLite, and options analysis covering Greeks, sentiment, and anomaly detection. It exposes tools for single and batch options search, semantic and date-range RAG queries, analysis, and CSV/chart exports, all orchestrated with LangGraph and optionally deployed as a FastAPI microservice. Dependencies include langchain, langgraph, langchain-openai, langchain-chroma, chromadb, fastapi, uvicorn, pandas, matplotlib, and tavily-python. Configuration is via a .env file holding required OpenAI and Polygon.io API keys and optional Tavily and LangChain tracing keys, validated through a settings module — standard credential setup with no exfiltration described.
The target users are developers, quantitative analysts, and fintech builders who want to create conversational options-analysis assistants with RAG and multi-agent workflows. Use cases include querying options chains for specific tickers and expirations, batch-analyzing multiple symbols, semantically searching cached data, and exporting results to CSV or charts. As an investment-analysis tool it is informational and not financial advice.
An OpenAI API key and a Polygon.io API key are required, set in a .env file in the project root. Tavily (web search) and LangChain tracing keys are optional. A settings module validates the keys before the agent runs.
LangGraph for agent orchestration, ChromaDB for the RAG knowledge base and caching, SQLite for persistent conversation memory, and FastAPI/uvicorn for optional microservice deployment. It requires Python 3.10+.
Options analysis including Greeks, sentiment, and anomaly detection, plus single-ticker and batch options search with caching, semantic and date-range retrieval from the RAG store, and exports to CSV or charts.
Options search tools cache results in ChromaDB (with a force_refresh flag to bypass the cache), while conversation state is persisted across sessions via SQLite using a thread_id checkpoint, so a session can be resumed later.
No. It is a data-retrieval and analysis tool for options research; outputs are informational. Users are responsible for validating data and making their own trading decisions, and for the cost of the paid APIs it calls.
Quick Setup:
.claude/skills/Repository
aradotso/data-skills