Data Formulator

What it does

Data Formulator takes in data from CSV, TSV, Excel and JSON files, screenshots, pasted text, local folders, databases and platforms such as Databricks. From that data it produces tables, natural-language explanations, editable charts and shareable reports, all inside one visual workspace.

People load data, ask questions in plain language, and review the answers in a Data Thread, a single history of questions, tables and charts. You can edit a result directly, branch from any earlier step to try an alternative analysis, or hand the next investigation to the AI analyst. Data connectors and a data memory help the agents track how different sources relate. Charts come from the Flint visualization language, and finished analyses can be composed into a report.

It suits analysts and developers comfortable running a Python package, a Docker container or a desktop build, who want exploratory analysis without writing code for every chart. It is a Microsoft Research project released under the MIT license. Running it requires access to a language model, and the page lists OpenAI, Azure, Ollama and Anthropic through LiteLLM. The Windows and macOS desktop builds are unsigned automated previews, and version 0.8 is currently a beta.

What it can do

• Load CSV, TSV, Excel, JSON, screenshots and text, or connect to databases and data platforms
• Ask natural-language questions and get explanations, tables and editable charts
• Branch alternative analyses from any earlier step in the Data Thread
• Refine 30+ chart types built on the Flint visualization language
• Compose analysis results into shareable reports
• Use persistent sessions and workspaces with an English/Chinese interface

Works with

• OpenAI
• Azure
• Ollama
• Anthropic via LiteLLM
• Databricks
• DuckDB

Example workflows

Explore a spreadsheet

Input: a CSV or Excel file. Steps: upload it, ask a question in natural language, review the resulting chart, then branch to a follow-up question. Output: a set of related charts in one Data Thread.

Build a shareable report

Input: data from a connected database. Steps: have the analyst load the data, ask several questions, refine the charts. Output: a report composed from the results.

Before you choose

• Needs a configured language model (OpenAI, Azure, Ollama or Anthropic), which may carry separate costs
• Desktop builds are automated previews, not code-signed or notarized, and trigger OS warnings
• Version 0.8 is in beta; the stable release is 0.7

Listed under

• Data
• Analysts
• Data analysis
• Databases
• Data exploration
• Data visualization
• Open data

Get started

Run uvx data_formulator or pip install data_formulator, then python -m data_formulator.

1. Install with uvx data_formulator or pip install data_formulator, or download a desktop build
2. Launch it and open http://localhost:5567 in the browser
3. Configure a supported language model and load your data

The source code is available under MIT. Model-provider usage and compute are separate.

Source: github.com/microsoft/data-formulator, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• GitHub - microsoft/data-formulator: 🪄 Data Formulator is an interactive AI-powered data analysis system makes it easy to connect, explore and visualize data. · GitHub ↗github.com

Company website

data-formulator.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.