Hugging Face drops an open-source spreadsheet that runs thousands of AI models on your data—no code needed

Hugging Face just released AI Sheets, an open-source tool that drags the spreadsheet interface into the era of large language models. It’s a no-code playground for building, transforming, and enriching datasets using thousands of open models from the Hugging Face Hub. You can deploy it locally or run it for free on the Hub, and it even supports OpenAI’s gpt-oss. The pitch is dead simple: you start with a dataset or generate one from a plain English description, then create new columns by writing prompts that reference existing ones. The tool handles the API calls and model inference behind the scenes.

The real power move here is the feedback loop. You can manually edit or validate individual cells, and those corrected examples get automatically folded back into your prompts as few-shot demonstrations. It’s prompt engineering by way of a spreadsheet, letting you iterate fast on small samples before committing to costly, large-scale pipelines. Hugging Face calls it a “vibe test” for models, and that feels about right. You can spin up a column for each model you want to compare, then create a judge column to have an LLM score the outputs side-by-side.

Use cases range from the practical (cleaning messy text columns, classifying content, extracting zip codes with web search) to the speculative (generating synthetic professional bios and realistic emails for pharma executives). The synthetic data angle is particularly interesting for developers stuck without access to real data due to privacy constraints. You describe a persona, AI Sheets generates the bios, and then another column writes emails in that persona’s voice. It’s fast, it’s transparent, and because it’s open-source and local-first, your data stays under your control.

I’m cautiously optimistic here. The spreadsheet metaphor lowers the barrier for domain experts who know their data but don’t write Python. But the real test will be whether the prompt-to-cell feedback loop actually produces reliable results at scale, or if it’s just a slick demo that falls apart when you need consistency across 50,000 rows. For now, it’s a genuinely useful addition to the data wrangling toolkit—and a reminder that the most impactful AI tools often look boring on the surface.