PCRE2 and on-device planning · inspect before apply

Turn a plain-English edit into regex on Mac

ZapPad can use Apple’s on-device Foundation Models to propose a supported text transformation from a short instruction. It validates the proposed operation and shows a preview for review before you apply it. This is a helper, not a guarantee that a model understood your intent; test the preview.

Currently offered to US customers only · Apple Silicon (M1+) · macOS 26+ for AI generation; macOS 12+ for manual regex

Plain-English Regex & Reusable Macros on Mac | ZapPadPlain-English regex demo · 2:51

The embedded player loads only when you choose to play it. YouTube processes playback data under its own privacy policy. Watch on YouTube ↗

From instruction to a reviewed transformation

  1. Describe the exact change in a short instruction, such as masking part of an email address.
  2. ZapPad’s local planner proposes a supported regex or macro operation.
  3. Inspect the pattern and generated action, then test the preview on representative text.
  4. Apply only after the result matches your intent; save supported steps as a macro if you will repeat them.

Only bounded instructions and optional examples go to Apple’s on-device model. The full document stays in the local workflow. The model must be available on the Mac; there is no cloud fallback in the current build. Conventional regex search works without AI.

What the generated regex can and cannot do

ZapPad compiles PCRE2 patterns and validates supported replacement captures, but a pattern can be syntactically valid and still describe the wrong transformation. Review the match count and output, include edge cases, and keep an original or use the app’s recovery path.

The planner supports a bounded set of operations and does not act as a general-purpose coding agent. It does not read an entire file into a model context, and it does not infer that a destructive change is safe. Regex search for ordinary documents is distinct from literal search in the huge-file viewer.

Examples where review matters

Mask a value

Keep the first character and domain of each email address. Inspect matches that are malformed or contain plus tags.

Normalize identifiers

Capture a predictable ID shape and change only its prefix. Check that numbers and neighboring fields remain intact.

Repeat a known cleanup

After validating one transformation, save a supported macro and replay it locally on later text.

These examples illustrate review practice, not a promise that every free-form request maps to a supported operation.