Python library that turns scripts into interactive web apps for data dashboards, reports and chat apps.
- Stars
- 17,580
- Forks
- 1,729
- Open issues
- 117
- Last commit
- 21 hours ago
What is it?
Data Formulator is a visual workspace for exploring and analyzing data with AI agents. You can start from CSV, TSV, Excel, JSON, screenshots, text, or databases and data platforms such as Databricks. An analyst agent helps locate and load sources, proposes a loading plan, and lets you review data before it joins the workspace.
Instead of a long chat log, questions are organized in a Data Thread. Each step shows its reasoning, tables, and editable charts. You can refine a result directly, branch from any earlier step to compare an alternative, or hand off the next investigation to the agent. Finished analyses can be composed into a report.
Charts are built on Flint, an open-source visualization language that compiles compact chart specs into visualizations. The project runs locally via uvx or pip, or with Docker, and it also offers Windows and macOS desktop builds.
Key features
- 01Connects to CSV, Excel, JSON, databases, and data platforms like Databricks.
- 02AI analyst agent discovers sources and proposes a loading plan for review.
- 03Data Thread keeps each question, answer, and chart in one branchable history.
- 04Charts are editable and built on the Flint visualization language.
- 05Compose analysis results into shareable reports.
- 06Runs locally via uvx or pip, or with Docker.
Use cases
- Exploring a new CSV or Excel dataset with natural-language questions and charts.
- Comparing alternative analysis paths side by side without losing earlier context.
- Querying a company database or data platform through an agent and building charts from the results.
- Turning exploration results into a report to share with colleagues.
Quick start
- 1
Install and run via uv (recommended)
uvx data_formulator - 2
Or install via pip
pip install data_formulator - 3
Run after pip install and open the browser at localhost:5567
python -m data_formulator
Things to know
- Desktop preview builds for Windows and macOS are not code-signed or notarized, so the OS will show a warning.
- The latest version is a beta preview (0.8.0b1); the stable release is 0.7.
- AI features need a configured model provider, so you need your own API key or a local model.
Topics
FAQ
- What is data-formulator?
- AI-powered data exploration tool that connects to files and databases and turns questions into editable visualizations, from Microsoft Research. Data Formulator is a visual workspace for exploring and analyzing data with AI agents. You can start from CSV, TSV, Excel, JSON, screenshots, text, or databases and data platforms such as Databricks. An analyst agent helps locate and load sources, proposes a loading plan, and lets you review data before it joins the workspace.
- Is data-formulator free to use? What license does it have?
- data-formulator is open source under the MIT license. Check the license terms before using it in commercial products.
- Is data-formulator still maintained?
- Actively maintained. The last commit to data-formulator was on Oct 20, 2026.
- What can I use data-formulator for?
- Exploring a new CSV or Excel dataset with natural-language questions and charts. Comparing alternative analysis paths side by side without losing earlier context. Querying a company database or data platform through an agent and building charts from the results. Turning exploration results into a report to share with colleagues.
- How do I install data-formulator?
- Install and run via uv (recommended): uvx data_formulator Or install via pip: pip install data_formulator Run after pip install and open the browser at localhost:5567: python -m data_formulator
- Where can I get data-formulator?
- The source code and installation instructions are on GitHub: https://github.com/microsoft/data-formulator
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A drop-in Jupyter replacement with an AI agent, a `.deepnote` YAML notebook format, and a block-based editor for Python, R, and SQL.
Python package for quickly building and sharing web demos and UIs for machine learning models, APIs, or any Python function.
A 10-week, 20-lesson beginner curriculum from Microsoft covering core data science topics with Python, quizzes and projects.
A curated Awesome-style list of Python resources for data science, covering libraries, tutorials, books, datasets and talks.
A Python low-code toolkit for building multi-page data visualization dashboards built on Plotly, Dash and Pydantic.