wtf is datalyze?

Datalyze is a data analysis assistant that lets a user upload an Excel or CSV file and get automatic charts back, then keep refining those charts through natural language chat. A person drags in a spreadsheet, the tool immediately produces a pie chart and a line chart, and then typing something like look at profit by product or show quarterly trends updates the charts on the spot. Under the hood, a React frontend sends the uploaded file and chat messages to a FastAPI backend, which hands the request to an AI agent built with LangChain and powered by the DeepSeek language model. The agent uses four specific tools: one that reads the structure of the uploaded file, one that groups and aggregates the data, one that analyzes trends over time, and one that turns the results into a fixed JSON format the frontend can render directly as a chart. The language model decides which tools to call and in what order, but the actual number crunching is done by pandas inside those tools, so the model itself never touches the raw data directly. It works with CSV, XLSX, and XLS files and is not tied to any specific column names, so it can be pointed at sales data, car listings, real estate data, or other tables. The project ships with three sample datasets covering sales, car sales, and housing data so a user can try it immediately. Uploaded files are cached so a user can switch back to a previous file's results without re-uploading. Setting it up requires Python 3.10 or newer, Node.js 18 or newer, and a DeepSeek API key, since the chat and analysis are powered by that external model. The backend runs with uvicorn and the frontend with Vite, both locally. The project is released under the MIT license.