What this is + why it works — Technical dialect

Same primitives, register inverted from Serene. Geist + Instrument Serif + JetBrains Mono — but here Mono leans heaviest. It's a data tool. Mono says we care about your data and we show our work. Serif appears exactly once: as the pivot word structure in the hero.
Teal is the trust signal. Everywhere a value is "safe," everywhere we want the eye to rest after a warning — teal. The dot grid background uses teal at 4.5% so the field itself reads like "checked, clean, instrumented."
Purple shows up exactly once. On the download CTA. It's the do work energy from Hybrid — we earn it by reserving it for the conversion moment. Everything else stays in navy + teal + Mac-window grays.
Hero is a side-by-side, not a single column. Headline + lede on the left, terminal proof on the right. The terminal has the glass + mask-composite edge-light primitive — the only loud moment in the page, because it is the product.
Before/after CSV is the proof point, not a feature card. Two side-by-side panels with realistic-but-fake data: Jennifer Martinez → Denise Holloway, SSN 412-38-9201 → 738-52-1094. After-panel gets edge-light treatment. Same primitive as Hybrid, used for the moment that earns it.
Benefits live in a single mono-typed strip. Four cells, one border-grid, no glass — local-only / no uploads / works offline / structure preserved. Quieter than feature cards. The information is the design.
Detection types render as a chip cloud. 13 sensitive types, color-dotted by class — red for stripped, teal for safe-and-passed-through. Lets the buyer scan exactly what gets touched.
Download URL is real — GitHub. CleanSchema currently ships as a Streamlit app via git clone + run.command; the install card mirrors what the live site offers. Marked with the actual GitHub repo URL — no data-todo needed because this URL exists.

v1.2 · macOS local · no upload open source

Strip the secrets.
Keep the structure.

CleanSchema detects PII and sensitive values in your .csv and .xlsx files and replaces them with realistic synthetic data. Same column names, same data types, same row count — zero real values. Runs entirely on your Mac.

Download for Mac free $ see how it works

cleanschema · employees_q1.csv

→ load "employees_q1.csv" // 847 rows · 12 cols

✗ sensitive first_name NAME → synthesize

✗ sensitive email EMAIL → synthesize

✗ sensitive salary FINANCIAL → synthesize

✗ sensitive ssn ID → synthesize

✓ safe department CATEGORICAL → keep as-is

✓ safe hire_year NUMERIC → keep as-is

✓ done → "employees_q1_clean.csv" // 0 real values · safe to share

// proof

Same shape. Different values.

employees_q1.csv
847 rows · 12 columns
processed in 0.34s

employees_q1.csv // before sensitive

first_name	email	salary	ssn	dept
Jennifer	j.martinez@acme.com	$87,500	412-38-9201	Engineering
Marcus	m.chen@acme.com	$112,000	203-77-4501	Engineering
Priya	p.shah@acme.com	$95,800	596-12-8430	Design
David	d.wright@acme.com	$78,200	719-04-3322	Operations
Aisha	a.patel@acme.com	$103,400	831-66-9011	Engineering

5 of 847 rows shown // real PII · do not share

employees_q1_clean.csv // after safe to share

first_name	email	salary	ssn	dept
Denise	d.holloway@example.org	$91,200	738-52-1094	Engineering
Owen	o.bennett@example.org	$108,300	419-22-6783	Engineering
Camila	c.ortiz@example.org	$98,600	662-08-5142	Design
Nathan	n.briggs@example.org	$81,900	277-91-3805	Operations
Imani	i.kelley@example.org	$106,750	514-49-2278	Engineering

5 of 847 rows shown // 0 real values

Categorical fields like dept pass through untouched. Statistical distributions match the source. Joins still work.

// why it ships

Built for one job. Done in seconds.

01 — local

Stays on your Mac

No upload. No API call. No cloud processing. The file enters memory, leaves as a clean copy.

02 — offline

Air-gapped friendly

Works with Wi-Fi off. Suitable for restricted environments where data can't leave the network.

03 — preserved

Structure intact

Column names, data types, row counts, joins, distributions — all preserved. Only the values change.

04 — auditable

Open source

Every classification rule and replacement function is in the repo. Read it. Fork it. Trust it on inspection, not on faith.

// flow

Three steps. One clean file.

~30s from open to download
no configuration required

step 01

Drop your file

CSV or Excel. Any size, any structure. CleanSchema reads it locally — nothing transmitted, nothing logged.

$ drop employees_q1.csv → cleanschema

step 02

Review what was detected

13 sensitive-data classifiers run against column names and value patterns. You see every detection — and can override any of them — before anything is replaced.

$ 4 sensitive · 8 safe · review →

step 03

Download the clean copy

Sensitive values become realistic synthetics. Names look like names. Salaries fall in the same range. Joins still work because IDs replace consistently.

$ save employees_q1_clean.csv ✓

// the promise

Your data never leaves your Mac.

CleanSchema runs as a local Python app — no servers, no API keys, no telemetry. The clean file lands in the same folder. The original is untouched. There is nothing for us to see, because we never built the pipe to see it.

no internet connection required
no account, no login, no email
no analytics, no telemetry, no logs
open source — every line auditable
works in air-gapped environments

network requests0

data uploaded0 bytes

accounts requirednone

telemetry events0

third-party SDKs0

your file location~/wherever/it/was

// detection engine

Thirteen sensitive types. Detected on column name and value shape.

red ● = stripped
teal ● = passed through

Financialsalary, revenue, amount

Namefirst_name, full_name, customer

Emailemail, mail, contact

Phonephone, mobile, cell, tel

Identity / IDssn, employee_id, account_no

Addressstreet, address, city, state

Datedob, hire_date, created_at

ZIP / Postalzip, postal_code, zip_code

Free textnotes, description, comments

Categoricaldepartment, status, region

Numericcount, units, quantity

Percentagerate, pct, completion

Booleanactive, is_admin, enabled

Strip the secrets. Keep the structure.