What is AI Sheets?

Hugging Face has launched AI Sheets, a free, open-source, local-first no-code tool that brings LLM-powered dataset creation and enrichment to a familiar spreadsheet interface. The idea is simple: let users work with rows and columns while applying natural language prompts to generate, clean, or transform data using open-source models like Qwen, Kimi, Llama 3, and custom models that follow the OpenAI API spec.

Key features

• No-code spreadsheet UI that lets non-technical users interact with models through prompts and visual cells.
• Instant access to thousands of models from the Hugging Face Hub or local inference endpoints, enabling fine-tuned or domain-specific models without cloud dependency.
• Local-first deployment keeps data on your machine for privacy and compliance, while an in-browser Spaces version enables instant trials.
• Collaborative workflows for rapid prototyping, manual validation, and large-scale pipeline runs to generate or enrich datasets.

How it works

Columns and cells are driven by plain-language prompts. You create a column, provide a prompt, and the selected model generates or enriches cell values across rows. For local usage, set environment variables such as MODEL_ENDPOINT_URL and MODEL_ENDPOINT_NAME to connect AI Sheets to a local inference server (for example, Ollama running Llama 3). AI Sheets is compatible with services that implement the OpenAI API spec, so many local or third-party endpoints can be used.

Common use cases

• Rapid dataset enrichment: add labels, generate paraphrases, or expand short prompts into longer examples.
• Data cleaning and transformation: normalize text, extract fields, or reformat content at scale.
• Annotation and validation: collaboratively edit model outputs, correct mistakes, and refine prompts to improve quality.
• Batch generation and prototyping: run large pipelines to create synthetic examples for model training or evaluation.

Deployment and getting started

Try AI Sheets directly in the browser via Hugging Face Spaces for immediate experimentation. For privacy and performance, clone the repository from GitHub (huggingface/aisheets), configure your local inference endpoint, and run the tool within your infrastructure. The project provides documentation and blog posts with step-by-step setup instructions and example workflows.

Why it matters

AI Sheets lowers the barrier to advanced dataset workflows by combining the accessibility of spreadsheets with direct LLM integration. Data scientists can iterate faster, analysts can automate routine tasks, and non-technical team members can participate in dataset creation and quality control — all while maintaining the option to keep data local and private.