Hugging Face introduces spreadsheet-style LLM dataset tool.
Hugging Face has launched AI Sheets, an open source, no-code toolkit that lets users work with datasets using thousands of AI models. Designed to resemble a spreadsheet, the platform lowers barriers to AI-driven data management by allowing dataset creation, transformation, and enrichment without programming skills.
A spreadsheet built for AI data
AI Sheets presents itself like a traditional spreadsheet interface where users can create new columns by writing prompts, clean up existing datasets, or generate synthetic data. Hugging Face said the tool can run locally on a personal computer or be deployed through its Hub and Spaces, making it accessible to developers, educators, and researchers alike.
The toolkit supports a wide range of tasks, including model comparison, prompt testing, classification, analysis, and data enrichment. A feature allows LLMs to act as judges, evaluating or comparing responses side by side.
Lowering barriers to dataset work
By removing the coding requirement, Hugging Face aims to make dataset work more inclusive. The company said AI Sheets offers an entry point for non-technical users such as business analysts, teachers, or researchers who want to test or refine prompts, run comparisons, or clean data. Community responses on Reddit have already highlighted its simplicity and open-source availability.
Hugging Face, known for hosting one of the largest repositories of open models, said the release is aligned with its mission of “making AI tools more transparent and accessible.”
Practical uses across industries
Early examples show how AI Sheets can be applied across multiple domains:
- Testing and comparing models: Users can run the same dataset through several models, add columns for each output, and use another LLM to judge the results.
- Improving prompts: Datasets can be iteratively refined, with users editing or validating AI-generated cells. These edits become examples that guide future outputs.
- Transforming data: A column can be created to clean punctuation, translate content, or standardise entries.
- Classification and analysis: Large datasets can be categorised or summarised with prompt-based columns.
- Enrichment: Incomplete datasets, such as addresses missing zip codes, can be expanded with AI prompts, even enabling web search when needed.
- Synthetic dataset generation: From realistic emails to fictional company profiles, users can create datasets for testing or training purposes without exposing private information.
According to Hugging Face, the tool supports both quick experimentation and production-scale workflows, with the option to export finished datasets back to the Hub. The company has also released documentation and tutorials to guide adoption.
Integrated with the open source ecosystem
AI Sheets is tightly connected to Hugging Face’s broader open source infrastructure. Users can import data in formats like XLS, CSV, TSV, or Parquet, or generate new datasets from scratch by describing them in natural language. The tool works with thousands of open source models, including gpt-oss from OpenAI, and supports multiple inference providers.
The flexibility extends to workflow design: users can manually edit, regenerate, or ‘thumbs up’ results to give feedback. These corrections are then reused as few-shot examples, improving subsequent outputs.
Open source, not commercialised
The release comes at a time when open-source AI projects are gaining traction against proprietary systems. Hugging Face has not announced commercial plans for AI Sheets, underscoring its commitment to transparency and adaptability rather than monetisation.
The company said the platform is meant for both experimentation and production, bridging accessibility and real-world application in data projects.
