Skip to content
Get packs

Guide · Open source

How to use Data Formulator: install, explore data and build charts

Updated · written with AI from the project's own documentation and checked against our catalogue data

Data Formulator is an open-source, AI-powered data visualization app from Microsoft Research. You combine clicks in a visual workspace with plain-language questions answered by AI agents, and it produces charts from files, CSVs, screenshots, text or databases. This guide shows how to install it, open it in your browser and use Data Threads to follow several lines of analysis at once.

Data Formulatormicrosoft/data-formulator · ★ 17.6k · MIT

Quick start

  1. 1Install uv or pip. Choose uv for the fastest route or pip inside a virtual environment. Docker is only needed for the Docker option.
  2. 2Start the app. Run uvx data_formulator, or run pip install data_formulator followed by python -m data_formulator.
  3. 3Open the browser. Go to http://localhost:5567 if it does not open automatically.
  4. 4Load your data. Start from a file, CSV, screenshot, text or database connection.
  5. 5Ask a question. Describe what you want to see in plain language or click through the workspace.
  6. 6Branch in a Data Thread. Create a new branch for each follow-up so you can compare analysis paths.

What Data Formulator does

Data Formulator is a Microsoft Research project for exploring data with visualizations powered by AI agents. It lets you mix UI interactions with natural language, so you can state what you want to see, branch into alternative analyses and share results.

You can start from any data format the project supports, including screenshots, text, CSV files or databases.

  • Data connectors give agents a common way to connect to different data sources.
  • Data memory keeps track of the relationships between those sources.
  • Data Threads let you branch into different questions and compare paths without losing context.

Install Data Formulator with uv (recommended)

uv is a Python package manager. If you already have uv installed, you can run Data Formulator without a separate setup step. Run uvx data_formulator --help to see the available options, such as a custom port, sandboxing mode and data storage location.

uvx data_formulator

Install Data Formulator with pip

If you prefer pip, the README recommends installing into a virtual environment. Install the package, then run it as a Python module. Data Formulator opens in your browser automatically at http://localhost:5567.

pip install data_formulator # install python -m data_formulator # run

Run Data Formulator with Docker

If you use Docker, build and start the containers from the project directory. Then open http://localhost:5567 in your browser. To stop the app, press Ctrl+C or run docker compose down.

docker compose up --build

Explore data and follow up with questions

Start by loading the data you want to analyze. Then combine clicks in the workspace with plain-language questions. Each answer can lead to a follow-up, and Data Threads keep the path you took visible so you can return to earlier steps.

Use the desktop builds only if you accept the warnings described below. The README says the preview builds are not code-signed or notarized.

  • Pick the chart or question you want to build on.
  • Branch into a new question from an earlier result.
  • Compare threads side by side to see which analysis holds up.

Choose an AI model provider

The README lists OpenAI, Azure, Ollama and Anthropic as supported model providers, accessed through LiteLLM. The README does not detail a fully offline setup, so if your data cannot leave your network, check the provider options before you load anything sensitive.

What people use it for

  • Exploring a CSV dataset with charts and follow-up questions
  • Comparing several lines of analysis without losing track of earlier steps
  • Trying AI-assisted charting on local data through a browser interface
  • Reading or modifying an open-source AI data visualization app as a developer

Frequently asked questions

Is Data Formulator free to use?

Data Formulator is released under the MIT licence, as stated in the project's README. The MIT licence permits use of the open-source code. Check the model provider you connect to, since the README does not describe their pricing.

What do I need to install Data Formulator?

You need Python, though the README excerpt does not state a version. You also need uv or pip, depending on the install route, and Docker only if you use the Docker option.

How do I run Data Formulator?

Run uvx data_formulator, or install with pip install data_formulator and then run python -m data_formulator. Open http://localhost:5567 in your browser if it does not open automatically.

Can Data Formulator work offline?

The README says it supports models such as OpenAI, Azure, Ollama and Anthropic through LiteLLM, but it does not detail a fully offline setup. If you cannot send data to a model provider, verify the setup before using it.

Is Data Formulator a replacement for Apache Superset or Metabase?

Data Formulator focuses on AI-assisted exploration with charts and branching questions. Apache Superset and Metabase are BI platforms, and the README does not describe enterprise governance controls beyond mentioning governed data sources, so pick based on whether you need a governed production BI platform.

Why does macOS or Windows warn me when I open the desktop app?

The README says the preview desktop builds are not code-signed or notarized. Only bypass the warning if you downloaded the archive directly from the project's workflow artifacts or releases.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Data Formulator suits analysts who want to explore a dataset with AI-assisted charts and keep a clear trail of follow-up questions, and developers who want to read or change an open-source app. It is a poor fit for teams that need a fully governed production BI platform, or for anyone who cannot send data to an AI model provider.

Install options & downloads