Data Formulator: AI-powered Data Visualization

🪄 Explore data with visualizations, powered by AI agents.

 

       

News 🔥🔥🔥

[03-02-2026] Data Formulator 0.7 (alpha) — More charts, new experience, enterprise-ready

• 📊 30 chart types with a new semantic chart engine (area, streamgraph, candlestick, pie, radar, maps, and more).
• 💬 Hybrid chat + data thread — chat woven into the exploration timeline with lineage, previews, and reasoning.
• 🤖 Unified DataAgent replacing four separate agents, plus new recommendation and insight agents.
• 🏗️ Workspace / Data Lake — persistent, identity-based data management with local and Azure Blob backends.
• 🔒 Security hardening — code signing, sandboxed execution, authentication, and rate limiting.
• 📦 UV-first build — reproducible builds via uv.lock; uv sync + uv run data_formulator.
• 📝 Detailed writeup on the new architecture coming soon — stay tuned!

Previous Updates

Here are milestones that lead to the current design:

• v0.6 (Demo): Real-time insights from live data — connect to URLs and databases with automatic refresh
• uv support: Faster installation with uv — uvx data_formulator or uv pip install data_formulator
• v0.5.1 (Demo): Community data loaders, US Map & Pie Chart, editable reports, snappier UI
• v0.5: Vibe with your data, in control — agent mode, data extraction, reports
• v0.2.2 (Demo): Goal-driven exploration with agent recommendations and performance improvements
• v0.2.1.3/4 (Readme | Demo): External data loaders (MySQL, PostgreSQL, MSSQL, Azure Data Explorer, S3, Azure Blob)
• v0.2 (Demos): Large data support with DuckDB integration
• v0.1.7 (Demos): Dataset anchoring for cleaner workflows
• v0.1.6 (Demo): Multi-table support with automatic joins
• Model Support: OpenAI, Azure, Ollama, Anthropic via LiteLLM (feedback)
• Python Package: Easy local installation (try it)
• Visualization Challenges: Test your skills (challenges)
• Data Extraction: Parse data from images and text (demo)
• Initial Release: Blog | Video

View detailed update history

• [07-10-2025] Data Formulator 0.2.2: Start with an analysis goal

  • Some key frontend performance updates.
  • You can start your exploration with a goal, or, tab and see if the agent can recommend some good exploration ideas for you. Demo

• [05-13-2025] Data Formulator 0.2.1.3/4: External Data Loader

  • We introduced external data loader class to make import data easier. Readme and Demo
    • Current data loaders: MySQL, Azure Data Explorer (Kusto), Azure Blob and Amazon S3 (json, parquet, csv).
    • [07-01-2025] Updated with: Postgresql, mssql.
  • Call for action link:
    • Users: let us know which data source you'd like to load data from.
    • Developers: let's build more data loaders.

• [04-23-2025] Data Formulator 0.2: working with large data 📦📦📦

  • Explore large data by:
    1. Upload large data file to the local database (powered by DuckDB).
    2. Use drag-and-drop to specify charts, and Data Formulator dynamically fetches data from the database to create visualizations (with ⚡️⚡️⚡️ speeds).
    3. Work with AI agents: they generate SQL queries to transform the data to create rich visualizations!
    4. Anchor the result / follow up / create a new branch / join tables; let's dive deeper.
  • Checkout the demos at [https://github.com/microsoft/data-formulator/releases/tag/0.2]
  • Improved overall system performance, and enjoy the updated derive concept functionality.

• [03-20-2025] Data Formulator 0.1.7: Anchoring ⚓︎

  • Anchor an intermediate dataset, so that followup data analysis are built on top of the anchored data, not the original one.
  • Clean a data and work with only the cleaned data; create a subset from the original data or join multiple data, and then go from there. AI agents will be less likely to get confused and work faster. ⚡️⚡️
  • Check out the demos at [https://github.com/microsoft/data-formulator/releases/tag/0.1.7]
  • Don't forget to update Data Formulator to test it out!

• [02-20-2025] Data Formulator 0.1.6 released!

  • Now supports working with multiple datasets at once! Tell Data Formulator which data tables you would like to use in the encoding shelf, and it will figure out how to join the tables to create a visualization to answer your question. 🪄
  • Checkout the demo at [https://github.com/microsoft/data-formulator/releases/tag/0.1.6].
  • Update your Data Formulator to the latest version to play with the new features.

• [02-12-2025] More models supported now!

  • Now supports OpenAI, Azure, Ollama, and Anthropic models (and more powered by LiteLLM);
  • Models with strong code generation and instruction following capabilities are recommended (gpt-4o, claude-3-5-sonnet etc.);
  • You can store API keys in .env to avoid typing them every time (copy .env.template to .env and fill in your keys).
  • Let us know which models you have good/bad experiences with, and what models you would like to see supported! [comment here]

• [11-07-2024] Minor fun update: data visualization challenges!

  • We added a few visualization challenges with the sample datasets. Can you complete them all? [try them out!]
  • Comment in the issue when you did, or share your results/questions with others! [comment here]

• [10-11-2024] Data Formulator python package released!

  • You can now install Data Formulator using Python and run it locally, easily. [check it out].
  • Our Codespaces configuration is also updated for fast start up ⚡️. [try it now!]
  • New experimental feature: load an image or a messy text, and ask AI to parse and clean it for you(!). [demo]

• [10-01-2024] Initial release of Data Formulator, check out our [blog] and [video]!

Overview

Data Formulator is a Microsoft Research prototype for data exploration with visualizations powered by AI agents.

Data Formulator enables analysts to iteratively explore and visualize data. Started with data in any format (screenshot, text, csv, or database), users can work with AI agents with a novel blended interface that combines user interface interactions (UI) and natural language (NL) inputs to communicate their intents, control branching exploration directions, and create reports to share their insights.

Get Started

Play with Data Formulator with one of the following options:

• Option 1: Install via uv (recommended)

  uv is an extremely fast Python package manager. If you have uv installed, you can run Data Formulator directly without any setup:

  # Run data formulator directly (no install needed)
  uvx data_formulator

  Or install it in a project/virtual environment:

  # Install data_formulator
  uv pip install data_formulator
  
  # Run data formulator
  python -m data_formulator

  Data Formulator will be automatically opened in the browser at http://localhost:5567.

• Option 2: Install via pip

  Use pip for installation (recommend: install it in a virtual environment).

  # install data_formulator
  pip install data_formulator
  
  # Run data formulator with this command
  python -m data_formulator

  Data Formulator will be automatically opened in the browser at http://localhost:5567.

  you can specify the port number (e.g., 8080) by python -m data_formulator --port 8080 if the default port is occupied.

• Option 3: Codespaces (5 minutes)

  You can also run Data Formulator in Codespaces; we have everything pre-configured. For more details, see CODESPACES.md.

  

• Option 4: Working in the developer mode

  You can build Data Formulator locally if you prefer full control over your development environment and develop your own version on top. For detailed instructions, refer to DEVELOPMENT.md.

Using Data Formulator

Load Data

Besides uploading csv, tsv or xlsx files that contain structured data, you can ask Data Formulator to extract data from screenshots, text blocks or websites, or load data from databases use connectors. Then you are ready to explore.

Explore Data

There are four levels to explore data based depending on whether you want more vibe or more control:

• Level 1 (most control): Create charts with UI via drag-and-drop, if all fields to be visualized are already in the data.
• Level 2: Specify chart designs with natural language + NL. Describe how new fields should be visualized in your chart, AI will automatically transform data to realize the design.
• Level 3: Get recommendations: Ask AI agents to recommend charts directly from NL descriptions, or even directly ask for exploration ideas.
• Level 4 (most vibe): In agent mode, provide a high-level goal and let AI agents automatically plan and explore data in multiple turns. Exploration threads will be created automatically.

data-formulator-tutorial.mp4

• Level 5: In practice, leverage all of them to keep up with both vibe and control!

Create Reports

Use the report builder to compose a report of the style you like, based on selected charts. Then share the reports to others!

Developers' Guide

Follow the developers' instructions to build your new data analysis tools on top of Data Formulator.

Help wanted:

• Add more database connectors (#156)
• Scaling up messy data extractor: more document types and larger files.
• Adding more chart templates (e.g., maps).
• other ideas?

Research Papers

• Data Formulator 2: Iteratively Creating Rich Visualizations with AI

@article{wang2024dataformulator2iteratively,
      title={Data Formulator 2: Iteratively Creating Rich Visualizations with AI}, 
      author={Chenglong Wang and Bongshin Lee and Steven Drucker and Dan Marshall and Jianfeng Gao},
      year={2024},
      booktitle={ArXiv preprint arXiv:2408.16119},
}

• Data Formulator: AI-powered Concept-driven Visualization Authoring

@article{wang2023data,
  title={Data Formulator: AI-powered Concept-driven Visualization Authoring},
  author={Wang, Chenglong and Thompson, John and Lee, Bongshin},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  year={2023},
  publisher={IEEE}
}

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repositories using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.