Introducing AI Sheets: A No-Code Tool to Build, Enrich, and Transform Datasets with Open AI Models

Hugging Face has launched AI Sheets, a no-code, open-source tool designed to help users build, enrich, and transform datasets using AI models—without writing any code. The tool integrates seamlessly with the Hugging Face Hub and supports thousands of open models from the Hub, including OpenAI’s gpt-oss, via Inference Providers or local deployment. AI Sheets features a familiar spreadsheet-like interface, making it easy to experiment with small datasets before scaling up. Users can create new columns by writing simple prompts, then refine results through manual edits, validation, and feedback. These interactions are automatically used as few-shot examples to improve future outputs, enabling rapid iteration and better model performance. The tool is ideal for a variety of tasks. It can be used to compare different AI models by generating responses side by side and evaluating them with a judge model. It supports prompt refinement by letting users edit or validate cells in real time, effectively fine-tuning prompts on the fly. Users can also clean, classify, analyze, and enrich datasets—such as adding missing zip codes by enabling web search—or generate synthetic data for privacy-sensitive applications. Getting started is simple. Users can either generate a dataset from scratch by describing their desired data in natural language or import existing data in CSV, TSV, XLS, or Parquet format. The interface allows for easy expansion by dragging to add more rows, and supports manual input for small datasets. Once data is ready, users can export their dataset to the Hugging Face Hub. This creates a reusable config file that can be used with HF Jobs to generate larger datasets at scale. For example, a script can run a pipeline using the same prompts and configurations, enabling reproducible and scalable data generation. AI Sheets is already being used in practical workflows. One example involves testing multiple models on app descriptions and using a judge model to compare their outputs. Another shows how to categorize questions from a Hub dataset using an LLM, with the ability to improve results by editing and validating initial outputs. A third example uses LLMs to evaluate the quality of generated images and descriptions. The tool is available to try for free directly in the browser at huggingface.co/spaces/aisheets/sheets, or users can install and run it locally from the GitHub repository. For advanced use, a PRO subscription unlocks 20x more inference usage. AI Sheets is a powerful addition to the open AI ecosystem, empowering researchers, developers, and data scientists to work with AI-driven data workflows efficiently and collaboratively.