Runcell is an AI agent extension designed specifically for Jupyter notebooks. It runs inside JupyterLab and can automatically write Python code, execute cells, debug errors, and explain data analysis results in real time. Its core positioning is "an AI agent that works in Jupyter." Unlike AI code editors that only provide code completion or single-turn suggestions, Runcell advances tasks around the complete notebook workflow: understanding context, planning steps, running code, checking output, and continuing to the next step based on actual results.
Main capabilities
-
From question to executed workflow: After the user describes the desired result, Runcell plans the notebook steps, writes Python code, executes cells, and fixes errors along the way. It supports multi-step workflows, automatically completing steps such as checks, transformations, analysis, and validation.
-
Understanding surrounding cells: Users can ask contextual questions about a transformation, result, or error, and Runcell reads the surrounding cells and outputs, applies fixes, and keeps the analysis moving forward.
-
Runnable experimental exploration: Users can try different analysis methods side by side, obtain real notebook outputs, and then decide, based on evidence, the direction suited to the problem and data. Conceptual explanations are also presented in the form of runnable cells.
-
Reading output and continuing results: Runcell can interpret tables, charts, statistics, and other cell outputs, so that subsequent decisions are based on the results actually produced rather than guesses based only on code. It also maintains context across iterations, keeping questions, decisions, and previous outputs connected, so that the next step builds on existing work.
How to use
Runcell runs as a JupyterLab extension, so users can use it without leaving Jupyter. The basic process is: ask a question or describe the desired result; Runcell examines the notebook context, data, and existing code; breaks the task down into notebook steps; writes and runs Python code; reads the output; and then continues advancing the analysis or answers the next question. According to official information, Runcell is free to start using, runs in the user's own JupyterLab, and does not require an API key. Specific details such as installation methods, whether it supports your own OpenAI or Anthropic API keys, whether it remembers projects across sessions, whether it supports long-running multi-step tasks, whether it can understand chart image outputs, whether it is compatible with existing notebooks, and whether there is a free plan are not yet provided in the official materials.Full description.
Use cases
Runcell is aimed at analysts, data scientists, and researchers who work in a notebook-driven way. Its work output needs to go beyond a single prompt or a single cell to form reproducible code, clear evidence, and interpretable results. Typical scenarios include: starting from a business or operational problem and obtaining reproducible code and interpretable results; exploring data, testing methods, building models as needed, and comparing results without manually assembling every notebook step; analyzing markets, portfolios, forecasts, and risk in risk and quantitative finance, and obtaining diagnostic information that can support conclusions; in research and experimentation, turning problems into code, charts, and reproducible evidence while keeping the analysis in Jupyter.
Differences from similar tools
Runcell emphasizes that it is a notebook-native agent. AI IDEs are usually built around files and source code, whereas Runcell executes cells, reads their output, and continues working based on what the code actually produces. Autocomplete predicts the next line, while Runcell turns a problem into a multi-step notebook workflow, runs it, diagnoses the results, and moves the work forward. Notebook chat can explain a cell, while Runcell takes action on the notebook—editing code, running cells, reading output, and continuing the analysis from the evidence.
Limitations
The official materials do not provide detailed information about pricing plans, specific installation commands, platform compatibility, supported models, data privacy, or performance limitations; this content is not yet available.
