microsoft/data-formulator

Data Formulator is a tool from Microsoft Research that helps people explore and visualize data using AI agents. The problem it tackles is that turning raw data, a spreadsheet, a screenshot, a chunk of text, or the contents of a database, into useful charts and reports normally takes a fair bit of fiddling with software. Data Formulator lets you start from data in almost any of those shapes and then talk to AI agents to build visualizations and reports around it. The interface blends two ways of working: clicking and dragging in the user interface, and writing requests in natural language. You can ask visualization questions, edit charts directly, or hand exploration tasks off to an agent that keeps track of what you have been doing across the conversation. The latest release adds persistent connections to outside data sources including Superset, Kusto, Cosmos DB, MySQL, PostgreSQL, MSSQL, S3, Azure Blob and BigQuery, plus single sign-on and sessions that survive restarts. The chart engine supports more than 30 chart types, area, streamgraph, candlestick, pie, radar and maps among them, and a separate style-refinement agent can restyle a rough chart through plain-English instructions about typography, color, layout and annotations. You would use Data Formulator if you want to explore a dataset or build a shareable report without writing visualization code by hand. It is a Python package installable through pip or the faster uv installer, and also runs via Docker or GitHub Codespaces, it connects to AI models from OpenAI, Azure, Ollama and Anthropic through a library called LiteLLM. The full README is longer than what was provided.