nblibram

The notebook Swiss Army knife—query, edit, and sanitize Jupyter files from the shell. Treats notebooks as structured documents so agents and humans can slice, mutate, and clean them without a GUI. Secret detection is powered by gitleaks (222+ built-in rules for API keys, tokens, credentials, etc.). Ships as a single Go binary with no runtime dependencies.

Philosophy

Notebooks are computational narratives—code cells don't stand alone, they're explained by the Markdown cells around them. nblibram's query operations slice Markdown/code cell pairs together so the surrounding story always travels with the code. section slices by heading hierarchy, cells slices by cell order, both preserving the narrative context that makes code understandable.

Commands (stdin → stdout)

All commands read .ipynb JSON from stdin (or --file) and write to stdout. Query commands apply gitleaks-based privacy filters by default (--no-filter to disable).

Query

• nblibram toc – heading structure with preview (--words, --format md|json).
• nblibram section – cells under a heading, including nested subsections (--sets, --format md|json|py).
• nblibram cells – consecutive cells from a matched position (--count N), or Markdown+code pairs (--sets N).
• nblibram outputs – cell outputs (--format text|json|raw, --mime).

Mutate

• nblibram insert – insert a cell (--query, --position before|after, --type code|markdown, --source).
• nblibram update – replace cell content (--query, --source, --hash required).
• nblibram delete – remove a cell (--query, --hash required).

Mutation commands write the modified notebook JSON to stdout. Use -i for in-place file update. --hash enforces optimistic locking—obtain it via nblibram hash.

Filter & Audit

• nblibram filter – sanitize sensitive information in a notebook (-i for in-place).
• nblibram audit – check for leaked secrets without modifying the notebook. Exits 1 if leaks are found (--format text|json).

Both use gitleaks' built-in rules by default. Set NBLIBRAM_GITLEAKS_CONFIG to a .gitleaks.toml path to add custom rules (e.g. IP addresses, domain names, internal URLs).

Utility

• nblibram hash – compute djb2 hashes for cells (--query or --all).
• nblibram pkl – read pickled kernel output logs (--file, --format json|text).

Queries

Use --query TYPE:VALUE to locate cells. Multiple --query flags are ANDed.

• start:37 – absolute cell index.
• match:"pattern" – regex against cell content.
• contains:"text" – substring match.
• id:abc123 – Jupyter cell_id.
• meme:UUID – nblineage meme ID. Trailing * for prefix match (e.g. meme:642f96e0* matches branched cells).

Examples

# Table of contents
nblibram toc --format json < notebook.ipynb

# Extract a section
nblibram section --query match:"## Setup" --sets 2 < notebook.ipynb

# Get 5 consecutive cells starting from index 10
nblibram cells --query start:10 --count 5 --format py < notebook.ipynb

# Extract output as PNG
nblibram outputs --query id:plot-cell --format raw --mime image/png < notebook.ipynb > plot.png

# Insert a cell, then sanitize
nblibram insert --query start:0 --source 'x = 1' < notebook.ipynb | nblibram filter > clean.ipynb

# Delete with optimistic locking
HASH=$(nblibram hash --query start:3 < notebook.ipynb | jq -r '.[0]._hash')
nblibram delete --query start:3 --hash "$HASH" -i notebook.ipynb

# Audit for secrets (CI-friendly, exits 1 on findings)
nblibram audit < notebook.ipynb

# Audit with JSON output
nblibram audit --format json < notebook.ipynb

# Read a pickled kernel log
nblibram pkl --file output.pkl --format text

Getting started

Install with pip

pip install nblibram

Wheels include the executable and install the nblibram command in your Python environment. Go is not required.

Standalone binary

Download the latest binary from Releases and place it in your $PATH:

# macOS (Apple Silicon)
curl -sL https://github.com/NII-cloud-operation/nblibram/releases/latest/download/nblibram_darwin_arm64.tar.gz | tar xz
mv nblibram /usr/local/bin/

# macOS (Intel)
curl -sL https://github.com/NII-cloud-operation/nblibram/releases/latest/download/nblibram_darwin_amd64.tar.gz | tar xz
mv nblibram /usr/local/bin/

# Linux (amd64)
curl -sL https://github.com/NII-cloud-operation/nblibram/releases/latest/download/nblibram_linux_amd64.tar.gz | tar xz
sudo mv nblibram /usr/local/bin/

Build from source

go build ./cmd/nblibram/

Configuration

nblibram uses gitleaks for secret detection. By default, gitleaks' 222+ built-in rules are active.

To add custom rules (e.g. IP addresses, internal URLs), create a .gitleaks.toml and point to it:

export NBLIBRAM_GITLEAKS_CONFIG=/path/to/.gitleaks.toml

Example .gitleaks.toml:

[extend]
useDefault = true

[[rules]]
id = "ipv4-address"
description = "Detects IPv4 addresses"
regex = '''\b\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}\b'''
[rules.allowlist]
regexes = [
    '''192\.168\.\d{1,3}\.\d{1,3}''',
    '''127\.0\.0\.1''',
    '''0\.0\.0\.0''',
]

[[rules]]
id = "email-address"
description = "Detects email addresses"
regex = '''[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}'''
[rules.allowlist]
regexes = [
    '''.*@example\.com''',
]

Detected secrets are replaced with [rule-id_N] labels, preserving equivalence (e.g. the same secret always maps to [generic-api-key_1]).