nblite

Notebook-driven Python development made simple.

nblite is a tool for developing Python packages using Jupyter notebooks, and/or plaintext notebooks (e.g. percent format .pct.py files) as the source of truth. Write your code in notebooks, add export directives, and nblite generates clean Python modules automatically.

Note: nblite was inspired by the excellent nbdev, with some adjustments to make it more lightweight and quality-of-life additions. Full credit of the concept and implementation of notebook-driven development using Jupyter notebooks should go to the creators of nbdev.

Features

• Notebook-first development: Write code in Jupyter notebooks with full interactivity
• Automatic module generation: Export marked cells to Python modules
• Multiple formats: Support for .ipynb, percent-style .pct.py, and standard .py modules
• Smart execution: Fill notebook outputs with parallel execution and change detection
• Git integration: Pre-commit hooks for auto-cleaning and validation
• Documentation generation: Build docs with MkDocs, Jupyter Book, or Quarto
• Flexible pipelines: Define custom export pipelines between code locations

Installation

pip install nblite

Quick Start

1. Initialize a project

mkdir myproject && cd myproject
nbl init --name mylib

This creates:

myproject/
├── nblite.toml     # Configuration file
├── nbs/            # Notebooks directory
└── mylib/          # Python package
    └── __init__.py

2. Create a notebook

nbl new nbs/core.ipynb --title "Core Module"

3. Add code with export directives

In your notebook, mark cells for export:

#|default_exp core

#|export
def greet(name: str) -> str:
    """Return a greeting message."""
    return f"Hello, {name}!"

#|export
class Calculator:
    """A simple calculator."""

    def add(self, a: int, b: int) -> int:
        return a + b

4. Export to Python modules

nbl export

This generates mylib/core.py:

# AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/core.ipynb

__all__ = ['greet', 'Calculator']

def greet(name: str) -> str:
    """Return a greeting message."""
    return f"Hello, {name}!"

class Calculator:
    """A simple calculator."""

    def add(self, a: int, b: int) -> int:
        return a + b

5. Fill notebook outputs

nbl fill

Executes all notebooks and saves their outputs. Uses smart change detection to skip unchanged notebooks.

Configuration

nblite is configured via nblite.toml:

# Export pipeline: notebooks -> modules
export_pipeline = "nbs -> lib"

# Code locations
[cl.nbs]
path = "nbs"
format = "ipynb"

[cl.lib]
path = "mylib"
format = "module"

# Git hooks (optional)
[git]
auto_clean = true
auto_export = true

# Notebook cleaning (optional)
[clean]
remove_outputs = false
remove_execution_counts = false

See the Configuration Guide for all options.

Export Directives

Control what gets exported with special comments:

See the Directives Reference for all directives.

CLI Commands

Use nbl <command> --help for detailed options.

See the CLI Reference for complete documentation.

Multi-Stage Pipelines

Define complex export pipelines:

# notebooks -> percent scripts -> modules
export_pipeline = """
nbs -> pcts
pcts -> lib
"""

[cl.nbs]
path = "nbs"
format = "ipynb"

[cl.pcts]
path = "pcts"
format = "percent"

[cl.lib]
path = "mylib"
format = "module"

This creates an intermediate representation in percent format, useful for:

• Code review (percent files are plain Python)
• Debugging export issues
• Version control of notebook content

Git Integration

Install git hooks for automatic cleaning and export:

nbl install-hooks

The pre-commit hook will:

1. Clean notebooks (remove outputs if configured)
2. Run the export pipeline
3. Validate staging state

Configure in nblite.toml:

[git]
auto_clean = true      # Clean notebooks before commit
auto_export = true     # Run export on commit
validate_staging = true # Warn about staging issues

Documentation Generation

Generate documentation from notebooks:

# Build documentation
nbl render-docs

# Preview with live reload
nbl preview-docs

Supported generators:

• MkDocs (default): pip install mkdocs mkdocs-material mkdocs-jupyter
• Jupyter Book: pip install jupyter-book
• Quarto: Install from https://quarto.org/

Configure in nblite.toml:

docs_cl = "nbs"           # Code location to document
docs_title = "My Project" # Documentation title
docs_generator = "mkdocs" # Generator to use

[docs]
output_folder = "_docs"

Notebook Execution

Fill notebooks with outputs:

# Execute all notebooks
nbl fill

# Execute specific notebooks
nbl fill nbs/core.ipynb nbs/utils.ipynb

# Parallel execution
nbl fill --workers 8

# Test without saving (dry run)
nbl test

Control execution with directives:

#|eval: false
# This cell is skipped during execution
expensive_computation()

#|skip_evals
# All following cells are skipped
...

#|skip_evals_stop
# Execution resumes here

Converting Existing Code

Convert Python modules to notebooks:

# Single file
nbl from-module utils.py nbs/utils.ipynb

# Entire directory
nbl from-module src/ nbs/ --recursive

Project Structure

A typical nblite project:

myproject/
├── nblite.toml          # Configuration
├── nbs/                 # Source notebooks
│   ├── 00_index.ipynb
│   ├── 01_core.ipynb
│   └── 02_utils.ipynb
├── mylib/               # Generated Python package
│   ├── __init__.py
│   ├── core.py
│   └── utils.py
├── _docs/               # Generated documentation
└── README.md            # Generated from notebook

Documentation

• Getting Started - Tutorial for new users
• Configuration Guide - Complete nblite.toml reference
• CLI Reference - All commands and options
• Directives Reference - Notebook directives
• Export Pipeline - How export works
• Git Integration - Hooks and workflows
• Documentation Generation - Building docs

Philosophy

nblite follows the literate programming philosophy: code and documentation live together. Notebooks are the source of truth, and Python modules are generated artifacts.

Key principles:

1. Notebooks first: Write and test code interactively
2. Explicit exports: Only marked cells become part of your library
3. Clean separation: Keep exploration separate from production code
4. Reproducibility: Fill outputs to ensure notebooks run correctly

Contributing

Contributions are welcome! Please see our contributing guidelines.

License

MIT License - see LICENSE file for details.

Acknowledgments

nblite is inspired by nbdev from fast.ai, reimagined as a lightweight, focused tool for notebook-driven development.