Dataset builder disguised as a spreadsheet

What it does AI Sheets is a web interface that turns the spreadsheet grid into a no-code control panel for synthetic data. You define columns that call language models—drawn from the Hugging Face Hub or from your own OpenAI-compatible endpoint—to generate, rewrite, or annotate rows. When the browser UI is too small for the job, you can offload bulk generation to Hugging Face Jobs.

The interesting bit Instead of exporting CSVs to a Python script, you configure inference directly in the sheet: pick a provider, set a prompt template, and let the cells populate themselves. The project is built on Qwik, which keeps the frontend reactive without the usual pile of client-side baggage.

Key highlights

• Works with thousands of Hub models out of the box, but can be repointed to local Ollama or any other OpenAI-compatible endpoint.
• Scales beyond the UI with batch scripts that run through Hugging Face Jobs.
• Can enrich data with web search results if a Serper API key is provided.
• Text-to-image generation is available, though it is locked to the Hugging Face Inference Providers API regardless of custom endpoint settings.
• Telemetry is enabled by default.

Caveats

• Text-to-image generation cannot be rerouted to custom endpoints; it always phones home to Hugging Face Inference Providers.
• Web search is disabled unless you bring your own Serper API key.
• Telemetry is opt-out rather than opt-in.

Verdict This is for data teams and prompt engineers who want to prototype synthetic datasets without writing glue code first. If you already have a parameterized Python pipeline, a spreadsheet UI is probably not the missing piece.