Data Formulator is an interactive AI-powered data analysis system developed by Microsoft Research that enables users to connect, explore, and visualize data through a unified visual canvas. Written primarily in TypeScript, the system addresses a core problem in data analysis: data lives across multiple platforms including databases, warehouses, BI tools, and files, yet traditional coding agents require extensive setup and return results as difficult-to-parse code or text. Data Formulator simplifies this workflow by allowing users to connect any data source, ask questions in natural language, and receive interactive charts that can be edited, branched, and shared without writing code.
The platform supports a comprehensive set of data connectors including databases like MySQL, PostgreSQL, and MSSQL, cloud data warehouses such as BigQuery and Azure Data Explorer, BI systems like Superset, and object storage services including S3 and Azure Blob Storage. Users can also extract data from Excel files, images, websites, and text blocks. The system features a data-loading agent that identifies and retrieves tables from connected databases, and a unified Data Agent with thread memory that inspects data, runs sandboxed code, and provides explanations and recommendations grounded in user context. The Data Thread interface keeps questions, intermediate results, and charts navigable, allowing users to revisit earlier steps, branch into alternative analyses, and compare results side by side.
The visualization capabilities include over thirty chart types spanning area charts, streamgraphs, candlestick charts, radar charts, maps, and KPI displays, powered by a semantic chart engine. A style-refinement agent transforms rough charts into presentation-ready visuals through natural language instructions. Users can build reports and export them as images or PDFs. The platform supports persistent sessions and workspaces that are identity-isolated and saved across restarts, and offers multilingual UI support with English and Chinese currently available.
According to GitGenius activity tracking, the repository shows strong engagement with a median issue and pull request response latency of 4.3 hours across 82 tracked items, and a mean latency of 33.6 hours. The most active contributor is Chenglong-MS with 190 tracked events, followed by Archer456 with 12 events and greentownwinwinwin with 8 events. The most frequently applied issue labels are help wanted with 5 instances, good first issue with 3 instances, and enhancement with 2 instances, indicating active community engagement and openness to contributions. The repository overlaps with contributors from langgenius/dify, ollama/ollama, and cline/cline, suggesting connections to broader AI and development tool ecosystems.
Installation is available through multiple methods including pip, uv package manager, Docker, GitHub Codespaces, and local development setup. The system supports multiple language models through LiteLLM integration, including OpenAI, Azure, Ollama, and Anthropic. The project is published under the MIT license and includes research papers documenting the underlying concepts and design, with the most recent paper addressing iterative visualization creation with AI agents.
