PandasAI

PandasAI is an open-source Python library (MIT-licensed) that lets you query CSVs, parquet files, SQL warehouses and pandas DataFrames in plain English. It wraps an LLM around a code-execution sandbox so a prompt like 'plot revenue by region for Q3' becomes real pandas/matplotlib code, runs it in an isolated Docker environment, and returns the dataframe, chart or summary. v3 supports multi-dataframe joins, semantic layers, and a LiteLLM backend so you can swap between OpenAI, Anthropic, local Ollama models or anything else LiteLLM supports.

It's aimed at two crowds: data analysts who want to skip boilerplate ('describe this dataset', 'find outliers'), and product teams embedding a chat-with-your-data feature without building the agent loop from scratch. The core library is free; the company behind it (Sinaptik) sells a managed cloud dashboard and a self-hosted enterprise edition (separate license, sales-led) that adds a BI-style UI, user management, and governed connectors on top of the OSS engine.

The Docker sandbox is the meaningful differentiator over rolling your own LangChain pandas agent — generated code can't escape into your host environment. Caveats: it's Python 3.8-3.11 only, results are only as good as the underlying LLM, and the 'AI Dashboard' marketing site is much thinner than the GitHub repo, which is where the real documentation lives.