AI DocPrep / for legal & compliance

Use AI on client documents without breaching confidentiality.

ABA Formal Opinion 512 expects you to protect client data before it reaches an AI tool. AI DocPrep redacts privileged details on your own machine, so files never leave the firm and no vendor keeps a copy.

runs on-prem·no vendor logging·MIT open source
your machine · offline
engagement-letter.docx
Client: Jane Whitfield SSN: 402-19-8823 Matter: 2026-CV-00841
redacted on this machine
engagement-letter.md
Client: NAME SSN: SSN Matter: 2026-CV-00841
internet requests during conversion: 0
the short version

Redaction that never trusts a third party with the original.

A cloud "redaction" service still receives the unredacted file. AI DocPrep runs entirely on the machine the file already lives on, so privileged material is never transmitted, cached, or used to train someone else's model.

why firms use it

Built for work that can't leave the building.

01

Your duty of confidentiality, intact

Strip client names, SSNs, account numbers, and identifiers before a single word reaches an AI tool — the step ABA 512 points to.

02

Nothing logged by a vendor

No account, no upload, no telemetry. The file never leaves the device, so there's no third-party copy to subpoena or breach.

03

Your IT team can vet it

The code is public under the MIT license. Security can read exactly what it does and confirm it makes no network calls, before approval.

how it redacts

Three levels of thoroughness, all local.

Match the depth to the sensitivity of the matter:

  • Instant patterns. SSNs, account and card numbers, emails, phone numbers, credentials.
  • On-device AI. A bundled model catches party names, firms, and places.
  • Local LLM. The deepest, context-aware pass, through your own Ollama server.
get it

Try it on one matter this afternoon.

Free while it's in beta — no account, nothing to pay. Signed and ready to run on macOS and Windows.