Data Formulator

Data Formulator is a Microsoft Research prototype that blends natural language and visual interfaces for agentic data exploration and visualization. Analysts can start from data in any format — CSV, images, text, or databases — and explore collaboratively with AI agents through a hybrid UI-and-NL interface. The platform offers an agent mode for fully delegated exploration, data-threads for branching workflows, and a multi-modal chart builder combining drag-and-drop with natural language. Version 0.7 added persistent data connectors to Superset, Kusto, Cosmos DB, MySQL, PostgreSQL, MSSQL, S3, Azure Blob, BigQuery, and others, with SSO, lazy catalog loading, and smart filtering.

Data Formulator’s conversational agent maintains thread memory, unifying explanation, exploration, visualization, and recommendation into fluid conversations that carry context across turns. Recent additions include a chart style-refinement agent that turns rough charts into presentation-ready visuals via natural-language tweaks to typography, color, layout, and annotations. Experimental knowledge-distillation work lets agents extract reusable skills from sessions into a shared library that informs future sessions, advancing the state of agentic data exploration as a research area.