Data-Copilot

Why we included this project

Data-Copilot is a research system that takes a natural-language data request and lets an LLM drive the whole job to completion. Instead of calling a fixed set of queries, the agent designs the interface functions it needs, then invokes them in sequence or in parallel to fetch and clean data, run analyses and predictions, and render charts. If you want to see how tool design, intent detection, and multi-step scheduling can be wired together with a model, the bundled code is organized cleanly enough to read and extend. It ships set up for Chinese stock, fund, and economic data through Tushare, and the model and data layers are separated, so swapping in your own sources or moving between GPT-3.5 and Qwen is manageable. Treat it as a reference implementation of end-to-end autonomous data analysis rather than a production analytics platform.