What it does Data Formulator is a browser-based data exploration tool that connects to databases, warehouses, files, and even screenshots, then lets you query and visualize through natural language. It runs locally via Python (uvx data_formulator or pip) and opens at localhost:5567. The latest v0.7 adds persistent workspaces, a semantic chart engine with 30+ chart types, and report export to image or PDF.

The interesting bit The “Data Thread” keeps your exploration navigable: you can revisit earlier steps, branch into alternative analyses, and compare side by side. That’s the hard part of AI-assisted analysis — not generating one chart, but not losing your place when you iterate. The style-refinement agent also lets you polish rough charts through natural language rather than fiddling with Vega-Lite by hand.

Key highlights

• Connectors for Superset, Kusto, Cosmos DB, MySQL, PostgreSQL, MSSQL, BigQuery, S3, Azure Blob, and a plugin system for custom sources
• Data extraction from images, text, websites, and Excel via a loading agent
• Sandboxed code execution with thread memory for context-aware exploration
• Multilingual UI (English and Chinese, more via contributions)
• Model-agnostic: OpenAI, Azure, Ollama, Anthropic through LiteLLM

Caveats

• Requires bringing your own LLM API keys; no built-in model
• v0.7 is fresh (May 2026), so the connector and workspace features are still settling
• Microsoft CLA required for contributions

Verdict Worth a spin for data teams tired of pasting between Jupyter, BI tools, and ChatGPT. Not for you if you need fully governed enterprise permissions or prefer writing D3 by hand.