dispose 0.1.0

Local privacy classifier and redactor for sensitive files (emails, text, markdown, PDFs)

dispose

Local privacy classifier and redactor for sensitive files — text, markdown, email (.eml), and PDF.

Built for scrubbing iCloud / personal email exports before you feed them to an AI agent. Sensitive spans are replaced with [REDACTED]. Nothing is sent to OpenAI or any other network service.

Source: codeberg.org/hiddendispatches/dispose

Install

Requires Python 3.9+.

pip install dispose

Then:

dispose --version
dispose categories

From source (contributors)

git clone https://codeberg.org/hiddendispatches/dispose.git
cd dispose
pip install -e ".[dev]"
pytest

Eight privacy categories

Same taxonomy as OpenAI Privacy Filter (without using their model or tool):

Category	What gets redacted
account_number	SSNs, credit cards (Luhn), bank accounts, routing numbers, IBANs, labeled account / member / policy IDs
private_address	Street addresses, city/state/ZIP, PO boxes
private_email	Email addresses
private_person	Names in email headers, salutations, signatures
private_phone	Phone numbers
private_url	URLs and IP addresses (including iCloud / share links)
private_date	Dates, especially DOB-labeled dates
secret	Passwords, API keys, tokens, JWTs, PEM private keys

Classifiers are deterministic pattern engines tuned for high recall on structured PII in correspondence. They run entirely offline.

Usage

# List the eight categories
dispose categories

# Read an .eml: decode base64 / quoted-printable so you can actually read it
dispose view path/to/mail.eml
dispose decode path/to/mail.eml          # alias for view
dispose view mail.eml --pager            # open in less/more
dispose view mail.eml -o readable.txt    # save decoded text

# Read decoded content AND redact privacy spans
dispose view mail.eml --redact
dispose view mail.eml --redact -o clean.txt

# Classify a file (report spans, do not write)
dispose classify path/to/email.eml
dispose classify ./exports --recursive --json

# Redact → writes path.redacted.ext by default
dispose redact path/to/notes.md
dispose redact ./icloud-export --recursive

# Print redacted text to stdout (pipe into an agent)
dispose redact message.eml --stdout

# Pipe / stdin
cat message.eml | dispose view -
cat message.eml | dispose redact -

# Overwrite in place (careful)
dispose redact sensitive.txt --in-place

# Only certain categories
dispose redact mail.eml -c private_email -c private_phone -c account_number

# Labeled placeholders: [REDACTED:private_email]
dispose redact mail.eml --label-categories

# OpenAI Privacy Filter-style alias
dispose filter mail.eml --stdout

Supported inputs

• Text / markdown: .txt, .md, .markdown, .rst, .log, …
• Email: .eml (MIME-parsed into headers + body by default; base64 and quoted-printable transfer encodings are decoded before redaction, including multipart bodies, nested messages, text attachments, and RFC 2047 headers)
• Structured text: .csv, .json, .yaml, .html, …
• PDF: .pdf (text is extracted, redacted, and written as a clean text PDF or plain text if you choose a non-.pdf output path)

Library use

from dispose import PrivacyPipeline

pipeline = PrivacyPipeline()
result = pipeline.redact(open("mail.eml").read())
print(result.redacted)
print(result.summary())
for span in result.spans:
    print(span.category, span.text, span.score)

Safety notes

• dispose is a data minimization aid, not a legal anonymization guarantee.
• Pattern classifiers can miss novel formats and can over-redact lookalikes.
• Review high-sensitivity outputs before sharing, even with an AI agent you trust.
• Prefer --stdout or a separate output directory over --in-place until you trust the result.

Releasing (maintainers)

Version lives only in pyproject.toml. Tag on Codeberg matches that version:

# after bumping version in pyproject.toml
git tag -a v0.1.0 -m "dispose 0.1.0"
git push origin main --tags

pip install -e ".[dev]"
pytest
python -m build
python -m twine check dist/*
# TestPyPI, then PyPI (API token; username __token__)
twine upload --repository testpypi dist/*
twine upload dist/*

PyPI does not require GitHub; Codeberg is the source host. Publish with a PyPI API token + twine (Trusted Publishing from Codeberg is not available yet).

License

MIT — see LICENSE.

Contributions

If you found this package useful, please consider donating or supporting development as a solo founder I've been without gainful employment for years.

Donations

SOL AH7TXdVn2ZddT7DALxSQz97AJBvweBXZdvV5vAojqgBy
ZEC BNB Chain 0xF7fE0DD3165053bD87217BF3e09041e123736175
TON/GRAM UQAnfI9jJj1iI5nxRcc0jXjrFtg41Ik7XvskB77mLNNMKwwK

Dev Studio https://6-castles.xyz

Support dispose-dev AT beconfidential DOT co