Anonymize client financials before AI ever sees them.
Built by a practicing CPA for §7216 and ABA Op. 512 workflows. Point it at a folder of financial-statement PDFs; every name, address, EIN, and account number becomes a stable token — and every dollar amount stays exactly where it was.
Provably offline — the test suite runs with networking disabled. Every dollar amount preserved. Verified before release.
One-time purchase · Windows · financial-statement PDFs
Before / after
Real output from the pipeline — on synthetic statements.
Names, addresses, EINs, and bank & loan numbers become consistent tokens. The numbers do not move.
Sunrise Holdings LLC Balance Sheet — As of December 31, 2025 EIN: 12-3456789 1234 Kapiolani Blvd, Ste 200, Honolulu, HI 96814 1000 · Cash - First Hawaiian Bank #4471082956 125,430.18 1250 · Due from Makai Property Partners LP 30,000.00 2600 · Mortgage - Bank of Hawaii Loan #883412907 612,500.00 Prepared by Malia K. Fernandez, CPA
Entity E LLC Balance Sheet — As of December 31, 2025 EIN: [EIN-2] [ADDRESS-2], City 1, HI [ZIP] 1000 · Cash - Bank 2 #[ACCT-1] 125,430.18 1250 · Due from Entity A 30,000.00 2600 · Mortgage - Bank 1 Loan #[LOAN-1] 612,500.00 Prepared by Person 2, CPA
Tokens stay consistent across every file in the batch, so the intercompany line still ties: “Due from Entity A” = 30,000.00 here matches “Due to Entity E LLC” = 30,000.00 on the related income statement.
Why it's different
Three things this does that a redaction tool doesn't.
100% local
Runs entirely on your machine — the test suite runs with networking disabled, so the offline claim is something you can check, not just read.
Analytical integrity
Dollar amounts survive byte-for-byte and tokens stay consistent across a whole batch, so intercompany lines and totals still tie after anonymizing.
Verification pass
A hard assertion that no mapped original name survives in the text-based output — run before anything is marked releasable.
See it run
Wi-Fi off. PDFs in. Anonymized out.
A three-minute walkthrough — from detection table to the verification report — is on the way.
Why now
The rules caught up with how we actually work.
Anony is designed for these workflows. Read the sources yourself — we don't paraphrase the requirements.
- OPR Alert 2026-19
IRS Office of Professional Responsibility guidance (June 24, 2026) on practitioner use of generative AI and Circular 230 duties. Read the analysis →
- ABA Formal Op. 512
The ABA's paradigm opinion on generative AI in legal practice — including the duty to protect client information when using AI tools. Read the overview →
- IRC §7216
Federal penalties for unauthorized disclosure of tax return information — the reason a consent must name the specific vendor. Read the statute →
⚠︎ Where the guarantee stops
Text-based PDFs get the full guarantee. The verification pass re-reads the output and hard-asserts that no mapped original name remains, and that every currency amount survived exactly.
Scanned / image-only pages are best-effort. Redaction covers the detected OCR word boxes and removes the image pixels beneath, but the verification pass can't re-read image content to confirm nothing was missed. For scanned documents, inspect the redacted PDF before sharing.
This limitation is true of every anonymizer's guarantee — we'd rather say it out loud.
Waitlist
Lock in launch pricing.
Join the waitlist and get Anony for $99 at launch instead of $149. Email only — no spam, no account.
One-time purchase · Windows · launch pricing $99 for the waitlist, then $149.
Waitlist sign-up activates shortly — check back soon.
You're on the list — launch pricing locked.