Runcell: AI Agent for Jupyter Notebooks | hokai.io

Last updated: 2026-06-04

Runcell adds an AI agent to JupyterLab that writes, runs, and debugs code from natural language. Free (50 credits/month). Pro: $20/month with 500 credits.

Runcell is a JupyterLab extension and AI agent that writes, runs, and debugs Python code inside Jupyter notebooks from natural-language instructions. It reads actual notebook state including variable values and cell outputs, making it more accurate than general-purpose coding assistants for data science tasks. The free Hobby plan includes 50 credits/month; Pro costs $20/month for 500 credits and unlimited code completions. Works with JupyterLab 4.0 and newer only.

About Runcell

Runcell is a JupyterLab extension and Python package that adds an AI agent directly inside Jupyter notebooks, enabling data scientists and analysts to write, execute, and debug Python code using plain-English instructions without leaving the notebook environment. Unlike general-purpose coding assistants, Runcell reads the actual notebook state, including cell outputs, DataFrame structures, variable values, and execution order, giving its AI grounding in the real data rather than guessing from file contents alone. The platform organizes its capabilities into four distinct modes. Autonomous Agent Mode takes full control of the notebook: it plans a multi-step analysis task, writes each cell, runs it, reads the output, fixes any errors, installs missing packages, and retries until the goal is complete. Reasoning Mode provides a step-by-step problem-solving layer for complex refactors or statistical analyses that require thinking through intermediate steps before writing code. Interactive Learning Mode functions as an AI tutor, producing live-coded examples to explain concepts like the difference between K-means and DBSCAN clustering directly in notebook cells. Smart Edit Mode acts as a context-aware chat assistant for targeted code edits, chart generation, and result explanations. Runcell is purpose-built for data professionals who spend the majority of their day in Jupyter. Common workflows include exploratory data analysis, automated data cleaning, feature engineering, visualization generation, and stakeholder report summarization, all executed with a single natural-language instruction instead of multiple manual cell edits. Pricing is credit-based. The free Hobby plan includes 50 credits per month (plus 50 bonus credits during beta). The Pro plan costs $20/month and includes 500 credits per month plus unlimited code completions, suitable for daily analytical work. The extension requires JupyterLab 4.0 or newer and is installed via pip or conda. As a newer tool with limited public company information, Runcell occupies a narrow but defensible niche as the only production-grade AI agent built specifically for Jupyter, rather than adapted from a general-purpose coding assistant.

Pricing

Hobby (Free): 50 credits/month plus 50 bonus credits during beta. Pro: $20/month includes 500 credits/month and unlimited code completions.

Key Features

• Autonomous Agent Mode: Plans multi-step data tasks, writes and runs each cell, reads outputs, fixes errors, installs missing packages, and retries automatically until the analysis is complete.
• Notebook State Awareness: Reads actual cell outputs, variable values, and DataFrame structures in memory so AI suggestions are grounded in real data, not inferred from source code alone.
• Reasoning Mode: Step-by-step problem-solving layer that thinks through complex statistical refactors or multi-stage analyses before writing a single line of code, reducing incorrect first attempts.
• Interactive Learning Mode: AI tutor that generates live-coded notebook cells to explain algorithms or methods (e.g., comparing K-means vs. DBSCAN) with working examples runnable directly in the session.
• Smart Edit and Chart Generation: Context-aware chat panel that generates visualizations, explains cell outputs in plain English, writes targeted code patches, and inserts results directly into notebook cells.

Pros

• Purpose-built for Jupyter: reads real cell outputs and DataFrame contents, so its code suggestions reference actual data values rather than placeholder logic that general-purpose tools like Copilot produce.
• Autonomous Agent Mode handles complete end-to-end workflows including error recovery and package installation, cutting a 45-minute manual EDA session to under 10 minutes for standard datasets.
• Free tier at 50 credits/month is sufficient for lightweight weekly analysis tasks, and the $20/month Pro tier is significantly cheaper than enterprise data science tools like DataRobot or Databricks AI.

Cons

• Works exclusively with Jupyter notebooks (JupyterLab 4.0+), making it incompatible with VS Code notebooks, Google Colab, or script-based Python workflows that represent a large share of data science work.
• Credit-based pricing means heavy Autonomous Agent Mode sessions can exhaust 500 monthly Pro credits within a week, leaving analysts rate-limited for the rest of the month.
• Company is early-stage with limited public documentation, no confirmed funding, and a small community, raising questions about long-term support and feature velocity compared to GitHub Copilot or Cursor.

Frequently Asked Questions

What is Runcell and what does it do?

Runcell is a JupyterLab extension and Python package that adds an AI agent directly inside Jupyter notebooks. It generates and executes Python code from natural-language instructions, reads actual cell outputs and DataFrame structures in memory, and operates in four modes: Autonomous Agent (full workflow automation), Reasoning Mode (step-by-step analysis), Interactive Learning (AI tutoring with live code), and Smart Edit (context-aware code assistance). It is designed specifically for data scientists and analysts who work daily in Jupyter.

How much does Runcell cost?

Runcell offers a free Hobby plan that includes 50 credits per month, plus 50 bonus credits during the beta period. The Pro plan costs $20 per month and includes 500 credits per month along with unlimited code completions and priority support. Credits are consumed per AI action, with complex Autonomous Agent sessions using more credits than simple chat edits. Credits do not roll over between months.

What are the main features of Runcell?

Runcell's four core modes are Autonomous Agent Mode (plans, writes, runs, and debugs complete analysis workflows), Reasoning Mode (step-by-step problem solving for complex code), Interactive Learning Mode (AI tutor with runnable code examples), and Smart Edit Mode (chat for targeted edits and chart generation). All modes read actual notebook state, including live variable values and DataFrame contents, giving them grounding that general coding assistants lack.

Is Runcell free to use?

Yes, Runcell has a free Hobby plan with 50 credits per month plus 50 bonus credits during the ongoing beta period. The free plan provides access to all four AI modes but with limited monthly credit allocation. For data scientists running daily analyses, the 50 credits per month are typically sufficient for lightweight weekly tasks but not for heavy multi-step Autonomous Agent workflows. Upgrading to Pro at $20/month unlocks 500 credits and unlimited code completions.

What are the best alternatives to Runcell?

For Jupyter users, the main alternatives are Jupyter AI (the official open-source extension), GitHub Copilot (available inside JupyterLab with limited notebook context), and AIMQ. For data scientists open to leaving Jupyter, Cursor and Windsurf offer stronger multi-file coding assistance in their own IDEs. Runcell differentiates itself from all of these by reading actual notebook state, making it the most Jupyter-native option, though Jupyter AI provides a fully free alternative for cost-sensitive teams.

Who is Runcell best for?

Runcell is best for data scientists who spend the majority of their working day in JupyterLab running exploratory analyses, cleaning datasets, and generating visualizations. It is also well-suited to academic researchers who want AI assistance within reproducible notebook workflows, and to Python learners who benefit from the Interactive Learning Mode's live code explanations. Runcell is not suitable for data engineers or ML engineers working primarily in Python scripts, VS Code notebooks, or Databricks, as it only integrates with JupyterLab 4.0 and newer.

Does Runcell have an API?

Runcell is installed as a Python package and JupyterLab extension via pip or conda, and does not currently expose a public REST API for external programmatic access. Integration is entirely within the JupyterLab interface using the Jupyter kernel and nbclient for execution. The extension communicates with JupyterLab's real-time collaboration layer (Jupyter RTC) to read and modify notebook state. For documentation on setup and configuration, visit docs.runcell.dev.

Visit Runcell Official Website