noirdoc redacts PII before Claude Code

This is the kind of privacy layer that turns AI coding from a demo-friendly toy into something legal and security teams can actually tolerate.

• –The main value is not “better prompting,” it’s reducing the blast radius of customer data before it ever leaves the machine
• –Local detection plus reversible placeholders is the right tradeoff for developer workflows that need context without exposing raw PII
• –Shipping the same engine as plugin, CLI, and proxy is smart distribution: one redaction core, three deployment models
• –The hosted proxy angle matters for teams that cannot rely on every engineer to use the plugin correctly
• –It sits in a growing category of AI guardrails that block sensitive data at the edge instead of trying to clean up after the fact