Exploring Hugging Face AI Sheets: No‑Code LLM‑Powered Data Manipulation

Overview

AI Sheets is an open‑source application released by Hugging Face that offers a spreadsheet‑style interface driven by large language models (LLMs). It enables users to create, transform, and enrich datasets through natural‑language prompts, eliminating the need for custom code.

Key Features

Supports thousands of open‑source LLMs, including OpenAI’s gpt‑oss.

Natural‑language prompts generate new columns, clean text, classify records, fill missing values, or synthesize entire rows.

Manual cell edits are taken into account for subsequent model generations.

Multiple output columns can be produced simultaneously, each powered by a different model.

An optional “reviewer” column can run a separate LLM to score or rank the generated results.

Supported Workflows

Schema‑first generation : Describe the desired data schema in plain English; AI Sheets creates a brand‑new synthetic dataset. This is useful for rapid prototyping and generating training data.

Import‑and‑transform : Upload existing files (CSV, TSV, XLS, Parquet) and apply LLM‑based transformations at scale. Users are encouraged to experiment on a small sample before scaling to the full dataset.

Model Comparison & Review Column

The interface allows side‑by‑side comparison of different models. For each row, separate columns can be populated by, for example, Qwen3‑Coder and gpt‑oss. An additional column can invoke a third model that evaluates the outputs and returns a quality score, enabling automated benchmarking within the spreadsheet.

Self‑Hosting with Docker

AI Sheets can be deployed locally using Docker, which is useful for data‑privacy‑sensitive scenarios. A typical deployment workflow is:

# Clone the repository (URL provided on the Hugging Face Hub)
git clone https://github.com/huggingface/ai-sheets.git
cd ai-sheets

# Build the Docker image
docker build -t ai-sheets .

# Run the container, supplying a Hugging Face access token and mounting a data directory
docker run -d \
  -p 7860:7860 \
  -e HF_TOKEN=YOUR_HF_TOKEN \
  -v $(pwd)/data:/app/data \
  ai-sheets

The container starts a web UI accessible at http://localhost:7860. Detailed instructions are available in the “Deployment Guide” published in Hub Spaces.

Export and Integration

After processing, datasets can be exported directly back to the Hugging Face Hub. The export operation also generates a reusable configuration file (in JSON/YAML format) that can be consumed by Hugging Face Jobs or downstream pipelines, simplifying end‑to‑end automation.

Availability

AI Sheets is free to use on the Hugging Face Hub, requires no local installation for the hosted version, and its full source code is publicly available on GitHub for self‑hosted deployments.