Home

data_redactor wiki

data_redactor is a Ruby gem that redacts secrets and PII from strings, hashes, and JSON at C speed — 89 built-in patterns across 9 tags, custom patterns, name patterns, and drop-in integrations for Logger, Rails, Rack, and the major LLM clients.

This wiki holds the deep material. The README stays the focused entry point (install, quick start, the public API); the wiki is where the full catalogue, engine internals, and cookbooks live. The generated API reference is on GitHub Pages — this wiki does not duplicate it.

⭐ Redact before it reaches the model — RubyLLM Integration

If you use RubyLLM, data_redactor scrubs secrets and PII out of every prompt, system instruction, and tool result before it leaves your process — either per-call, or transparently across every request with one line at boot. This is the integration to reach for first when you're sending free text to an LLM. → RubyLLM Integration

Pages

• RubyLLM Integration ⭐ — per-call redaction and install! transparent mode for RubyLLM, with caveats.
• Pattern Catalogue — all 89 patterns, grouped by tag and country, with examples and boundary-protection notes.
• C Engine Internals — NFA → bytecode → lazy DFA, the v19 story, and how a scan actually runs. Links the full research log.
• Integration Guides — Logger, Rails, Rack, and the Claude/OpenAI LLM helpers. (RubyLLM has its own page above.)
• Custom Patterns & Name Patterns — the cookbook for add_pattern and name_pattern.
• Performance & Benchmark Methodology — how the numbers are measured and what they mean.
• FAQ — thread-safety, what's not redacted, the base64/URL limitation, and other common questions.

Where things live (source of truth)

Topic	Canonical source
Release history, per-version diffs	CHANGELOG.md
Why decisions were made (research narrative)	DONE.md · research log
Open work	TODO.md
Public method signatures, params, return values	YARD docs on GitHub Pages
Install the gem	rubygems.org/gems/data_redactor

This wiki links back to those rather than restating them — if a number or a signature here ever disagrees with the source, the source wins.