NotebookFlow turns a Jupyter notebook into a visual workflow without moving
execution out of your local environment. Arrange notebook cells as connected
nodes, inspect their data flow, and run the pipeline from the canvas beside
the notebook.
Get started
Install the extension and open an .ipynb file.
Select the NotebookFlow: Open Canvas button in the notebook toolbar, or
run that command from the Command Palette.
On the first run, approve installation of the released NotebookFlow engine.
The extension creates a private Python environment in VS Code's extension
storage and reuses it on later runs. The engine includes the scientific
Python runtime, so its first download can take several minutes.
The Microsoft Jupyter extension is installed automatically as an extension
dependency. The local engine requires Python 3.11, 3.12, or 3.13. Node.js,
uv, and a NotebookFlow source checkout are not required.
What it does
Builds a pipeline graph from cells across a Jupyter notebook.
Keeps canvas edits and notebook source synchronized.
Streams local execution progress and cell outputs back into the notebook.
Provides Ask, Compose, Explain, and node-synthesis tools with your own LLM
provider key.
Optionally opens and saves notebooks in a NotebookFlow Cloud account.
Local execution and privacy
Pipeline code runs in the local notebookflow-app engine launched by the
extension. Notebook contents and execution data are not sent to NotebookFlow
Cloud. Cloud sign-in and cloud notebook commands are optional. If you enable
an LLM feature, its request is sent directly to the provider selected in your
settings using the key stored in VS Code SecretStorage.
Settings
Setting
Purpose
notebookflow.pythonPath
Optional Python 3.11–3.13 executable for managed-engine setup.
notebookflow.enginePath
Development-only override for an engine checkout run with uv.
notebookflow.llm.provider
LLM provider used by AI-assisted commands.
notebookflow.llm.model
Optional provider model ID.
notebookflow.cloud.baseUrl
Hosted NotebookFlow origin used only by optional cloud commands.
Use NotebookFlow: Set LLM API Key to store a key; it is never written to
workspace settings.
Troubleshooting
Python was not found: install Python 3.11–3.13, restart VS Code, or set
notebookflow.pythonPath to its executable.
Engine installation or startup failed: open Output: Show Output
Channels and select NotebookFlow Engine. Check network access to PyPI
for the first installation.
The canvas command asks for a notebook: open an .ipynb file with the
Jupyter extension first.
Remote development: the extension and managed Python engine run on the
remote extension host, beside the notebook files.
