data-formulator

Data Formulator is an open-source project from Microsoft Research that provides an interactive workspace for data exploration and analysis powered by AI agents. It addresses two core challenges in data work: the difficulty of connecting disparate data sources and the complexity of evolving analytical questions. The system offers data connectors that give AI agents a unified way to access files, databases, warehouses, and BI tools while maintaining a memory of relationships between sources. Data Threads allow users to branch into different analytical paths, compare results, and discover insights without losing context. Key capabilities include: - Multi-format data ingestion: CSV, TSV, Excel, JSON, screenshots, text, and direct database connections - Natural language interaction with AI agents for data exploration - 30+ chart types powered by the Flint visualization language - Branching analysis threads for exploring alternative questions - Report generation for sharing results - Support for multiple AI models via LiteLLM (OpenAI, Azure, Ollama, Anthropic) - Desktop applications for Windows and macOS - Docker deployment option - Python package installation via pip or uv The project is actively developed with recent releases introducing unified workflows, expanded data source support, and enhanced charting capabilities. It is licensed under MIT and welcomes community contributions.