microsoft/data-formulator

What it solves

Data Formulator addresses the difficulty of exploring data across fragmented sources and the challenge of maintaining context when analysis evolves through a series of follow-up questions. It eliminates the need for manual data connection setup and prevents the loss of analytical history that typically occurs in long chat-based interfaces.

How it works

The system uses AI agents to connect to diverse data sources (files, databases, warehouses) and maintain a "data memory" of their relationships. It features a visual workspace called "Data Threads," which allows users to ask questions in natural language, view the agent's reasoning, and branch off into alternative analysis paths without losing previous context. Visualizations are generated using Flint, an open-source visualization language that converts compact specifications into polished charts.

Who it’s for

It is designed for data analysts who want to communicate their intent via natural language and UI interactions to discover insights and create reports from various data formats, including screenshots, text, CSVs, and databases.

Highlights

• Multi-source connectivity: Supports CSV, TSV, Excel, JSON, screenshots, text, and databases like MySQL, PostgreSQL, MSSQL, Azure Data Explorer, S3, and Databricks.
• Data Threads: A branching conversation history that lets users compare different analysis paths and refine results.
• Semantic Chart Engine: Provides over 30 chart types powered by Flint with a dedicated style-refinement agent.
• Broad Model Support: Integrates with OpenAI, Azure, Ollama, and Anthropic via LiteLLM.