Microsoft's Data Formulator is an AI‑agent‑driven workspace that lets analysts load, query, and visualize data from disparate sources while preserving context through a data memory and branching Data Threads. The tool connects to files, local folders, databases, and platforms such as Databricks via unified data connectors that maintain relationships between sources. Powered by Flint, an open‑source visualization language, it compiles compact chart specs into polished visuals and offers a semantic chart engine that supports over 30 chart types, which can be refined by a style‑refinement agent. Version 0.8 beta 1 (released 08‑15‑2026) introduces a single flow—load data, ask questions, review results, branch in the Data Thread—adds support for additional data sources, and expands Flint‑based charts with recommendations, themes, and styling tools. Earlier milestones show v0.7 (05‑28‑2026) enabling governed source connection, agent‑based loading, persistent sessions, multilingual (English/Chinese) UI, and report sharing. Model integration relies on LiteLLM to access OpenAI, Azure, Ollama, and Anthropic models. Installation options include pip, uvx (e.g., uvx data_formulator==0.8.0b1), Docker compose, or building from source, with the application serving on http://localhost:5567.