Experiment

Data Formulator

Data Formulator blends natural language and visual interfaces to help analysts explore and visualize data with AI agents. Starting with any data (clean or messy, small or large), analysts can easily describe their goals to AI agents and collaboratively explore data in different directions in parallel to discover new insights. With AI agents recommending exploration ideas, transforming data and presenting results in interactive visual interfaces, users can deep dive into data in control without coding expertise. Finally, users can ask agents to create reports grounded in the explorations to share their insights.

In many real-world workflows, analysts discover insights iteratively, requiring frequent changes to both the data and the chart, and it is only within reach for users with coding and analytics experiences. Data Formulator is an interactive agentic tool to make exploration and visualization easy.

On the one hand, Data Formulator offers an agent mode that lets users vibe with data by fully delegating planning and exploration to AI agents from a high-level description. Based on the instruction, AI agents automatically generate executable plans leveraging data and exploration contexts, transform and visualize data to derive insights, interpret the results, refine follow-up plans, and continue exploration. This provides users with a hands-off experience to quickly get inspiration.

Yet, without control, users can get lost in insights that may not match their interests or needs. To stay in control, Data Formulator users can leverage the interactive visual interface to inspect results, provide fine-grained visualization instructions, and manage branching exploration directions. Data Formulator combines drag-and-drop chart building UI with NL for users to specify what they want to see, so that users can precisely describe their objective without verbosity of describing the design purely in natural language as in chat interface. To support iteration, users can leverage data-threads to control how AI agents should leverage exploration contexts to backtrack, branch or follow up based on specified charts as opposed to only the latest results. Finally, the computation behind the exploration is surfaced to the user visually, allowing them to inspect data, charts, and computation concepts behind the scenes to interpret the results and provide feedback to agents if needed.

By combining agentic experience with interactive control, Data Formulator makes it especially valuable for fast-paced exploratory analysis to help analysts derive and share insights with control.

Key to this flexibility are two innovations:

  • Agent mode supporting fully automated data exploration with AI agents from only a high-level exploration goal;
  • A multi-modal chart builder, which combines shelf-style chart configuration with natural language input for simple yet fine-grained specification
  • Data threads, which track previous chart versions and allow users to fork, revisit, or refine past ideas—supporting non-linear workflows

Data Formulator supports diverse input data: data screenshots, messy text blocks, structured csv/tsv files, and database tables, and can generate reports grounded on exploration threads to share insights.

Besides loading structured data and database tables, you can extract data from screenshots or messy data and perform analysis on top:

Data Formulator is designed based on our insights from our user studies, where we find a flexible interface for iteration is essential to accommodate different people’s working styles, their preferences of UI interaction for chart specifications, and the necessity to surface computation concepts to help them understand agent actions.

For data scientists and analysts seeking a smoother, faster way to prototype and evolve visualizations, Data Formulator offers a compelling approach. Try it out now and take part in shaping the future of data analysis.