CSV-AI is an open-source Streamlit application for chatting with, summarizing, and analyzing CSV files using large language models. Published on GitHub under the MIT license with about 340 stars and updated in May 2026, the current version is a v2 rewrite with a modular architecture and a provider-agnostic AI stack.
Upload a CSV and CSV-AI profiles it, builds prompt context from the schema and samples, and then lets you work with it three ways. Chat gives schema- and sample-aware question answering with token streaming. Summarize produces a single-call structured overview of the file. Analyze places deterministic pandas statistics, including charts, missingness, and correlations, side by side with an LLM analyst narrative, so you can check the model’s story against the numbers. The app runs on Streamlit Cloud or a local laptop, loads settings from environment variables or Streamlit secrets, and cleanly separates services from the UI so a FastAPI layer can be added later.
Under the hood, CSV-AI builds a profiler, a sampler, and a prompt-context builder for every uploaded file, which is why its answers can reference actual column names, value ranges, and sample rows instead of generic advice. Configuration can come from environment variables, a local .env file, or Streamlit secrets, with sensible defaults such as a low temperature aimed at factual answers for quick local runs.
CSV-AI is free and open source under the MIT license, including the hosted demo on Streamlit. You supply your own model API keys, or point it at a local Ollama model to avoid API costs.
CSV-AI is for analysts, students, and small teams who receive spreadsheet data and want quick answers and overviews in plain language before doing deeper work in a notebook.