runcell (Jupyter AI agent) is an AI copilot built for notebooks. It understands your current kernel state—variables, DataFrame schemas, imports, outputs—and proposes the next best cell to move your work forward. With one command, it can draft code, execute it safely, validate the output, and automatically refine on errors. You get production‑quality cells (with comments and docstrings) and a clear audit trail of changes.

Whether you’re exploring data, transforming pipelines, or teaching with notebooks, runcell eliminates “search–copy–paste–debug” loops. It nudges work toward best practices (tests, assertions, checkpoints), and documents decisions for future you—or your teammates. Bring your own LLM keys, keep data in your environment, and control exactly what leaves your notebook.

Core capabilities

Natural‑language to code

Drafts new cells or refactors existing ones from plain‑English prompts.

Incorporates notebook context (imports, variables, DataFrame dtypes, shapes) to produce runnable code.

Run, validate, iterate

Executes proposed cells, inspects outputs/exceptions, and auto‑fixes common issues (missing imports, dtype mismatches, off‑by‑one, plotting errors).

Context‑aware assistance

Explain cell: summarizes what a cell does and why.

Suggest next step: proposes analysis steps, visualizations, or checks based on your artifacts.

Inline docs: inserts comments, docstrings, and markdown rationale.

Environment & reproducibility

Detects missing packages and (optionally) generates a safe install cell; can export requirements.txt or environment.yml.

Data & visualization helpers

Quick EDA: profiling, missing‑value maps, summary stats.

Transformations: joins, groupby, window ops, feature engineering.

Visuals: histograms, pairplots, line/bar charts, residual plots, interactive charts (e.g., Plotly).

Works smoothly with pandas, numpy, polars, matplotlib/plotly, and visualization helpers like pygwalker.