This notebook is part of the nbsphinx documentation:


Sphinx Setup

In the directory with your notebook files, run this command (assuming you have Sphinx installed already):

python3 -m sphinx.cmd.quickstart

Answer the questions that appear on the screen. In case of doubt, just press the <Return> key repeatedly to take the default values.

After that, there will be a few brand-new files in the current directory. You'll have to make a few changes to the file named You should at least check if this variable contains the right things:

extensions = [

For an example, see this project's file.

Once your is in place, edit the file named index.rst and add the file names of your notebooks (without the .ipynb extension) to the toctree directive. For an example, see this project's doc/index.rst file.

Alternatively, you can delete the file index.rst and replace it with your own notebook called index.ipynb which will serve as main page. In this case you can create the main toctree in index.ipynb.

Sphinx Configuration Values

All configuration values are described in the Sphinx documentation, here we mention only the ones which may be relevant in combination with nbsphinx.


Sphinx builds all potential source files (reST files, Jupyter notebooks, ...) that are in the source directory (including any sub-directories), whether you want to use them or not. If you want certain source files not to be built, specify them in exclude_patterns. For example, you might want to ignore source files in your build directory:

exclude_patterns = ['_build']

Note that the directory .ipynb_checkpoints is automatically added to exclude_patterns by nbsphinx.


This is the only required value. You have to add 'nbsphinx' to the list of extensions, otherwise it won't work.

Other interesting extensions are:


Default language for syntax highlighting in reST and Markdown cells, when no language is specified explicitly.

By default, this is 'python3', while Jupyter doesn't have a default language. Set highlight_language to 'none' to get the same behavior as in Jupyter:

highlight_language = 'none'

See also nbsphinx_codecell_lexer.


See Custom CSS and html_css_files.

By default, a .txt suffix is added to source files. This is only relevant if the chosen HTML theme supports source links and if html_show_sourcelink is True.

Jupyter notebooks with the suffix .ipynb.txt are normally not very useful, so if you want to avoid the additional suffix, set html_sourcelink_suffix to the empty string:

html_sourcelink_suffix = ''


latex_additional_files can be useful if you are using BibTeX files, see References.


The configuration value mathjax_config can be useful to enable Automatic Equation Numbering.


Warnings can be really helpful to detect small mistakes, and you should consider invoking Sphinx with the -W option, which turns warnings into errors. However, warnings can also be annoying, especially if you are fully aware of the "problem", but you simply don't care about it for some reason. In this case, you can use suppress_warnings to silence specific types of warnings.

If you want to suppress all warnings from nbsphinx, use this:

suppress_warnings = [

You can also be more specific:

suppress_warnings = [

nbsphinx Configuration Values


If True, the build process is continued even if an exception occurs.

See Ignoring Errors.


Default Pygments lexer for syntax highlighting in code cells. If available, this information is taken from the notebook metadata instead.

Please note that this is not the same as highlight_language, which is used for formatting code in Markdown cells!


See Custom Notebook Formats.


See Prolog and Epilog.


Whether to execute notebooks before conversion or not. Possible values: 'always', 'never', 'auto' (default).

See Explicitly Dis-/Enabling Notebook Execution.


Kernel arguments used when executing notebooks.

If you use Matplotlib for plots, this setting is recommended:

nbsphinx_execute_arguments = [
    "--InlineBackend.figure_formats={'svg', 'pdf'}",
    "--InlineBackend.rc={'figure.dpi': 96}",

If you don't use LaTeX/PDF output, you can drop the 'pdf' figure format.

See Configuring the Kernels.


Input prompt for code cells. %s is replaced by the execution count.

To get a prompt similar to the Classic Notebook, use

nbsphinx_input_prompt = 'In [%s]:'


Use a different kernel than stored in the notebook metadata, e.g.:

nbsphinx_kernel_name = 'python3'

See Configuring the Kernels.


Output prompt for code cells. %s is replaced by the execution count.

To get a prompt similar to the Classic Notebook, use

nbsphinx_output_prompt = 'Out[%s]:'


See Prolog and Epilog.


Width of input/output prompts (HTML only).

If a prompt is wider than that, it protrudes into the left margin.

Any CSS length can be specified.


Options for loading RequireJS. See nbsphinx_requirejs_path.


URL or local path to override the default URL for RequireJS.

If you use a local file, it should be located in a directory listed in html_static_path.

Set to empty string to disable loading RequireJS.


If the browser window is narrower than this, input/output prompts are on separate lines (HTML only).

Any CSS length can be specified.


A dictionary mapping from a document name (i.e. source file without suffix but with subdirectories) -- optionally containing wildcards -- to a thumbnail path to be used in a thumbnail gallery.

See Specifying Thumbnails.


Controls when a cell will time out. The timeout is given in seconds. Given -1, cells will never time out, which is also the default.

See Cell Execution Timeout.


Options for loading Jupyter widgets resources. See nbsphinx_widgets_path.


URL or local path to override the default URL for Jupyter widgets resources. See Interactive Widgets (HTML only)).

If you use a local file, it should be located in a directory listed in html_static_path.

For loading the widgets resources, RequireJS is needed, see nbsphinx_requirejs_path.

If nbsphinx_widgets_path is not specified, widgets resources are only loaded if at least one notebook actually uses widgets. If you are loading the relevant JavaScript code by some other means already, you can set this option to the empty string to avoid loading it a second time.

Running Sphinx

To create the HTML pages, use this command:

python3 -m sphinx <source-dir> <build-dir>

If you have many notebooks, you can do a parallel build by using the -j option:

python3 -m sphinx <source-dir> <build-dir> -j<number-of-processes>

For example, if your source files are in the current directory and you have 4 CPU cores, you can run this:

python3 -m sphinx . _build -j4

Afterwards, you can find the main HTML file in _build/index.html.

Subsequent builds will be faster, because only those source files which have changed will be re-built. To force re-building all source files, use the -E option.

Note By default, notebooks will be executed during the Sphinx build process only if they do not have any output cells stored. See [Controlling Notebook Execution](executing-notebooks.ipynb).

To create LaTeX output, use:

python3 -m sphinx <source-dir> <build-dir> -b latex

If you don't know how to create a PDF file from the LaTeX output, you should have a look at Latexmk (see also this tutorial).

Sphinx can automatically check if the links you are using are still valid. Just invoke it like this:

python3 -m sphinx <source-dir> <build-dir> -b linkcheck

Watching for Changes with sphinx-autobuild

If you think it's tedious to run the Sphinx build command again and again while you make changes to your notebooks, you'll be happy to hear that there is a way to avoid that: sphinx-autobuild!

It can be installed with

python3 -m pip install sphinx-autobuild --user

You can start auto-building your files with

python3 -m sphinx_autobuild <source-dir> <build-dir>

This will start a local webserver which will serve the generated HTML pages at http://localhost:8000/. Whenever you save changes in one of your notebooks, the appropriate HTML page(s) will be re-built and when finished, your browser view will be refreshed automagically. Neat!

You can also abuse this to auto-build the LaTeX output:

python3 -m sphinx_autobuild <source-dir> <build-dir> -b latex

However, to auto-build the final PDF file as well, you'll need an additional tool. Again, you can use latexmk for this (see above). Change to the build directory and run

latexmk -pdf -pvc

If your PDF viewer isn't opened because of LaTeX build errors, you can use the command line flag -f to force creating a PDF file.

Automatic Creation of HTML and PDF output on

There are two different methods, both of which are described below.

In both cases, you'll first have to create an account on and connect your GitLab/Github/Bitbucket/... account. Instead of connecting, you can also manually add any publicly available Git/Subversion/Mercurial/Bazaar/... repository.

After doing the steps described below, you only have to "push" to your repository, and the HTML pages and the PDF file of your stuff are automagically created on Awesome!

You can even have different versions of your stuff, just use Git tags and branches and select in the settings which of those should be created.

Note If you want to execute notebooks (see [Controlling Notebook Execution](executing-notebooks.ipynb)), you'll need to install the appropriate Jupyter kernel. In the examples below, the IPython kernel is installed from the packet `ipykernel`.

Using requirements.txt

  1. Create a file named .readthedocs.yml in the main directory of your repository with the following contents:

    version: 2
     formats: all
       version: 3
         - requirements: doc/requirements.txt
       system_packages: true

    For further options see

  2. Create a file named doc/requirements.txt (or whatever you chose in the previous step) containing the required pip packages:


    You can also install directly from Github et al., using a specific branch/tag/commit, e.g.


Using conda

  1. Create a file named .readthedocs.yml in the main directory of your repository with the following contents:

    version: 2
     formats: all
       file: doc/environment.yml

    For further options see

  2. Create a file named doc/environment.yml (or whatever you chose in the previous step) describing a conda environment like this:

       - conda-forge
       - python>=3
       - pandoc
       - ipykernel
       - pip
       - pip:
         - nbsphinx

    It is up to you if you want to install nbsphinx with conda or with pip (but note that the conda package might be outdated). And you can of course add further conda and pip packages. You can also install packages directly from Github et al., using a specific branch/tag/commit, e.g.

    - pip:
         - git+
Note The specification of the `conda-forge` channel is recommended because it tends to have more recent package versions than the default channel.

HTML Themes

The nbsphinx extension does not provide its own theme, you can use any of the available themes or create a custom one, if you feel like it.

The following (incomplete) list of themes contains up to three links for each theme:

  1. The documentation (or the official sample page) of this theme (if available; see also the documentation of the built-in Sphinx themes)
  2. How the nbsphinx documentation looks when using this theme
  3. How to enable this theme using either requirements.txt or readthedocs.yml and theme-specific settings (in some cases)

Sphinx's Built-In Themes

3rd-Party Themes

If you know of another Sphinx theme that should be included here, please open an issue on Github. An overview of many more themes can be found at

Using Notebooks with Git

Git is extremely useful for managing source code and it can and should also be used for managing Jupyter notebooks. There is one caveat, however: Notebooks can contain output cells with rich media like images, plots, sounds, HTML, JavaScript and many other types of bulky machine-created content. This can make it hard to work with Git efficiently, because changes in those bulky contents can completely obscure the more interesting human-made changes in text and source code. Working with multiple collaborators on a notebook can become very tedious because of this.

It is therefore highly recommended that you remove all outputs from your notebooks before committing changes to a Git repository (except for the reasons mentioned in Pre-Executing Notebooks).

If there are no output cells in a notebook, nbsphinx will by default execute the notebook, and the pages generated by Sphinx will therefore contain all the output cells. See Controlling Notebook Execution for how this behavior can be customized.

In the Jupyter Notebook application, you can manually clear all outputs by selecting "Cell" $\to$ "All Output" $\to$ "Clear" from the menu. In JupyterLab, the menu items are "Edit" $\to$ "Clear All Outputs".

There are several tools available to remove outputs from multiple files at once without having to open them separately. You can even include such a tool as "clean/smudge filters" into your Git workflow, which will strip the output cells automatically whenever a Git command is executed. For details, have a look at those links: