Hi HN, I'm Andrios, founder of hoop.dev (YC W21), we build runtime controls for agents. We just released Alcatraz.

We built it because our product is written in Go and we do real time Data Masking, but we were using MS Presidio (Python) for PII detection and it made connections slow.

There was a good discussion here a few months ago when OpenAI released their Privacy Filter: (https://news.ycombinator.com/item?id=47870901). A lot of questions were built into our design. Structured identifiers (credit cards, SSNs, etc) are deterministic and you can verify them with checksum. Free text PII like names and addresses is where you actually need a model.

Alcatraz started as our internal solution, but we wanted to make it available for everyone: in-process PII detection in Go. Currently 45 entity types across 12 countries (US and Brazil included). Benchmarks vs MS Presidio (lib to lib) about 13x faster on dense documents, 100x on smaller ones.

Still in development, lots to improve. Feedback wanted: github.com/hoophq/alcatraz

Edit 1: fix typo Python written as Phyton