Runcell
Runcell is a Jupyter-native AI agent that writes and runs Python inside JupyterLab, reads the tables and charts your cells produce, fixes errors, and carries multi-step analyses forward on the notebooks you already have.
What is Runcell?
Runcell is a Jupyter-native AI agent for data analysis, data science and research, published by Kanaries Data Inc. It installs as a JupyterLab extension with a single pip install runcell command and works on the .ipynb notebooks you already have, rather than asking you to move into a new editor.
The distinction it draws is with AI coding tools built around files and source code. Where an autocomplete predicts the next line and a notebook chat explains a cell, Runcell acts on the notebook: it inspects the data, the existing code and the question, plans the steps, writes the Python, executes the cells in the live kernel, reads what came back, and continues from that result. The publisher describes this as a four-stage loop: inspect the context, plan and execute, read the outputs, keep moving.
Reading outputs is central to the claim. The agent reasons about the tables, statistics, charts and image results the cells produce, so the next decision rests on what actually happened rather than on the code alone. It also carries context across iterations and, according to the site, across sessions: you can return to a multi-week project and ask what has been done so far instead of re-explaining the dataset.
Model access is included. You pick a provider in product, with OpenAI, Anthropic and Google named in the privacy policy, and no API key of your own is required; the plan you are on determines which advanced models you reach. Usage is metered in monthly credits.
Four fields are targeted explicitly: data analysis, data science, risk and quantitative finance, and research and experimentation. Alongside the extension there is a desktop application for macOS and Windows, a set of free browser utilities, and Runcell Science, an Apache-2.0 open-source research agent that the publisher presents as a separate project needing no Runcell account.
The boundaries are firm: JupyterLab 4.4.0 and above only, classic Jupyter Notebook excluded, Python 3.10 or higher, no offline or air-gapped use, and no public developer API.
What it does
- Inspects the notebook, the data and the existing code before deciding what the next step should be
- Writes Python and executes the cells directly in the live kernel
- Reads cell outputs, from tables and statistics to charts and image results, and reasons from them
- Diagnoses runtime errors and applies the fix without leaving JupyterLab
- Breaks a single question into a multi-step notebook workflow and runs it end to end
- Keeps questions, decisions and previous outputs connected across iterations and across sessions
- Explains unfamiliar algorithms and libraries with runnable examples inside the notebook
When to use Runcell / When not to
A quick filter to help you decide if Runcell is the right fit.
When to use Runcell
- Data analysts who need a business question turned into reproducible code and evidence they can explain
- Data scientists exploring datasets, testing methods and comparing models without hand-assembling every notebook step
- Quantitative finance and risk professionals working on markets, portfolios, forecasts and credit-scoring diagnostics
- Academic researchers and doctoral students who must leave behind a notebook a reviewer can rerun
- Domain experts in biotech, chemistry or genomics who think in problems rather than in Python
When not to use Runcell
- Teams still on the classic Jupyter Notebook interface, or on JupyterLab below 4.4.0, which the extension does not support
- Organisations needing offline or air-gapped AI, since a network path to Runcell's service and to an LLM is required
- Developers looking for a public API to embed the agent inside their own product
- Buyers with strict compliance requirements, as no GDPR statement, no DPA and no security certification is published
- Mobile-first users: there is no iOS or Android application, and the desktop build covers only macOS and Windows
How to use Runcell
A typical end-to-end flow, from setup to results.
- Check the prerequisites first: Python 3.10 or higher and JupyterLab 4.4.0 or higher, the classic Jupyter Notebook interface being unsupported
- Install the extension from PyPI with the command pip install runcell
- Inside a conda environment, activate the environment first and still install with pip rather than with conda install
- As alternatives, use uv pip install runcell, the JupyterLab extension manager, or a wheel file taken from the releases
- Restart JupyterLab completely: a partial restart is the single most common reason the extension fails to appear
- Open the right-hand sidebar of JupyterLab and click the runcell icon
- Sign in through an OAuth identity provider such as Google
- Open one of your existing .ipynb notebooks and describe in plain language the result you want
- Let the agent plan the steps, write the Python, run the cells and recover from errors, then review the outputs it produced
- Optionally download the desktop application for macOS or Windows instead of working from the sidebar
Pros & Cons
Pros
- Installs into the environment you already use with a single pip command, with no new editor to learn
- Actually executes the cells and reads their outputs instead of stopping at a code suggestion
- No LLM provider API key to supply, since model access is included in the plan
- A permanent free plan with monthly credits, which makes a trial on a real project possible
- The model provider is chosen by the user rather than imposed by the vendor
- Privacy Mode is on by default on paid plans, and raw notebooks are not uploaded by default
- Unusually detailed installation and troubleshooting documentation, down to conda, Docker, Apple Silicon and Windows cases
Cons
- No GDPR statement, no Data Processing Agreement and no security certification are published anywhere on the site
- No hosting country and no hosting region is disclosed in the privacy policy
- On the free plan, Privacy Mode is unavailable and data may be used to train or fine-tune models
- Subprocessors are described by category only, the policy itself stating that a current list will be posted later
- No postal address, no legal notice page and no about page exist on the site
- Scope is narrow: classic Jupyter Notebook, offline use and a developer API are all excluded
- The credit system makes real cost depend on consumption per feature and per token rather than on the headline price
Pricing & Plans
Runcell is offered on a freemium basis. A permanent free plan, named Hobby, provides 20 credits per month at no cost. The cheapest paid entry point is the Pro plan at USD 20.00 per month, or USD 180.00 when billed annually. Paid plans are charged in advance on the cadence selected at checkout, and no refund is granted for a partial billing period unless required by law.
- USD 0 per month - 20 credits per month
- limited AI execution credits and limited access to LLM models
- USD 20 per month or USD 180 per year - 500 credits per month
- access to all advanced models
- priority support and Privacy Mode
- USD 60 per month or USD 600 per year - everything in Pro plus 2
- 000 credits per month
- access to all models
- priority access to new features and a cost per credit 25% lower than Pro
- USD 200 per month or USD 2
- 000 per year - everything in Pro+ plus 10
- 000 credits per month
- advanced model access
- priority support and Privacy Mode
- USD 40 per month or USD 400 per year - everything in Pro plus 500 credits per seat per month
- centralised billing
- invite links
- a team usage dashboard and admin seat controls
Data, GDPR & hosting
A consolidated view of how Runcell handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site: not in the privacy policy dated 17 September 2025, not in the terms of service, not on any other page. No legal basis, no data-subject rights section, no Article 27 EU representative and no data protection officer are named, and no Data Processing Agreement is published or offered on request. No certification such as SOC 2 or ISO 27001 is claimed either. What the publisher does document are technical measures: TLS in transit, encryption at rest, role-based least-privilege staff access, and patching practices it describes as appropriate for a SaaS in early availability. Privacy questions are routed to a support address. Kanaries Data Inc. is a US company with no EU establishment shown. This is an absence of any stated compliance, not evidence of a breach.
Who owns the data?
The terms are explicit: you remain the owner of the content you submit, and Kanaries Data Inc. uses it only to operate the service and to comply with the law. You grant the publisher the rights it needs to host and process that content in order to deliver the service, and nothing beyond. Kanaries retains all rights in the service itself, its documentation and its trademarks, and grants you a limited, non-exclusive, revocable licence to use Runcell for personal or internal business purposes. You stay responsible for whatever you submit, prompts, files and generated outputs alike. By default, raw notebooks and files never leave your own machine.
Reuse rights
Nothing in the terms restricts what you do with the code, figures and analyses Runcell produces. You own your content, and the licence granted to the publisher is limited to hosting and processing it in order to run the service, so no permission has to be asked before reusing an output. The counterpart is stated plainly: you are responsible for evaluating model outputs and applying human judgement before acting on them. On the publisher's side, use of your data depends on the plan. Paid plans run with Privacy Mode on by default, so prompts, context and outputs are not used to train or fine-tune models. Free-plan data may be retained and used to improve Runcell, model training and fine-tuning included. Third-party model providers are instructed not to train on your data wherever their own controls allow it, although limited abuse and quality checks on their side remain possible.
Data retention & training
Hosting summary
The privacy policy describes the data path but names no hosting jurisdiction. By default your raw notebooks and files stay on your machine; only minimal context, such as a traceback, a short code fragment or column names, is sent to the publisher's cloud prompt orchestrator, which assembles the prompt and forwards it to the LLM provider you selected. Conversation content is stored on the publisher's servers only if you enable cloud sync; with sync off it stays on your device, and prompt context is processed ephemerally for the active request. Declared protections are TLS in transit, encryption at rest and role-based least-privilege staff access. Subprocessors are given by category only: cloud hosting and storage, LLM providers, product analytics such as Microsoft Clarity, payload-free error monitoring, and billing. No country, region or cloud provider is named for storage. As a technical observation rather than a vendor statement, the domain resolves to an IP geolocated in the United States on Amazon infrastructure and flagged as an anycast node, and the publisher is a US company.
Things to keep in mind
Risks and trade-offs to weigh before adopting Runcell.
- On the free plan your prompts, notebook context and model outputs may be retained and used to train or fine-tune models, Privacy Mode being a paid-plan feature
- The extension can read your notebooks, cells, outputs and local files, so sensitive or regulated datasets deserve a deliberate decision before it is installed
- The GDPR is never mentioned and no Data Processing Agreement is offered, which makes the tool hard to justify in a regulated procurement process
- No hosting country or region is disclosed, so you cannot verify where your prompts and conversations are processed
- Letting an agent write, run and interpret an analysis can erode the habit of checking assumptions yourself, and the terms place that judgement squarely on you
- Cloud conversation sync stores your exchanges on the publisher's servers; it is optional, but it is a setting to check rather than to assume
- Liability is capped at USD 100 or twelve months of fees, and the abuse-fee clause still contains an unfilled placeholder amount
Setup & Integrations
Technical difficulty
Low for anyone already running JupyterLab: one pip install runcell command, a full restart of JupyterLab and an OAuth sign-in, quoted at about five minutes. The friction is environmental rather than conceptual. The two documented pitfalls are a partial restart, after which the extension simply does not appear, and Jupyter and runcell installed in different Python environments. Prerequisites are firm: Python 3.10 or higher and JupyterLab 4.4.0 or higher. The documentation covers conda, Docker, Apple Silicon, Windows and Linux cases, and a desktop build avoids the command line entirely.
Deployment
Integrations
Behind Runcell
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Runcell.
Frequently asked questions
What exactly is Runcell?
Which Jupyter versions does it support?
Do I need my own OpenAI or Anthropic API key?
Is there a free plan, and how much do paid plans cost?
Will my data be used to train AI models?
Does Runcell remember a project between sessions?
Can I use Runcell offline?
Is there a mobile app or a developer API?
What is the minimum age to use Runcell?
Should you pick Runcell?
Runcell is a deliberately narrow tool, and that is its strength. It does not try to be a general assistant: it targets the moment where a question has to become executed, inspectable notebook work, and it addresses that moment inside the place where the work already happens. For anyone who opens JupyterLab every day, the adoption cost is close to nothing, one pip command, a restart and a sign-in, and the free Hobby plan makes a genuine trial possible before any commitment. At USD 20 per month, the Pro entry point is modest for a tool that includes access to advanced models without asking for an API key.
The real differentiator is execution. AI editors suggest code; Runcell runs the cells, reads the tables and charts they produce, diagnoses the errors and continues from the evidence. Combined with cross-session memory and the conveniences it adds inside Jupyter, file tree, global search and git, it fits multi-step analytical work rather than isolated snippets.
The reservations are documentary rather than functional. The site publishes no GDPR statement, no Data Processing Agreement, no security certification, no hosting country and no postal address, and its subprocessors are described by category only. On the free plan, Privacy Mode is unavailable and data may be used to train models, which matters for anyone testing with real datasets. The publisher itself calls its security practices those of a SaaS in early availability, and the domain was only registered in June 2025, even though Kanaries Data Inc. stands behind established projects such as PyGWalker and RATH.
For an individual analyst or researcher working on non-sensitive data, Runcell is an easy and well-priced recommendation. For a regulated organisation, the legal paperwork will have to be requested directly before adoption.
- Choosing a selection results in a full page refresh.
- Opens in a new window.