Alcatraz: Go-native PII detection that outpaces MS Presidio
The motivation behind it is relatable. They were building runtime data masking for agents (their core product), but relied on Presidio for PII detection. Since Presidio is Python-based and their stack is entirely Go, every call introduced noticeable latency. Real-time systems don't tolerate that kind of friction.
What I found interesting is how they approached the problem. Structured identifiers like credit cards or SSNs are deterministic — you can validate them with checksums. But free-text PII like names or addresses needs a model. Alcatraz handles both, supporting 45 entity types across 12 countries including the US and Brazil. That's a solid range out of the gate.
On benchmarks, they're reporting ~13x speedup on dense documents and up to 100x on smaller ones when compared library-to-library with Presidio. Those numbers hold up if you're running in-process Go — no subprocess overhead, no serialization costs. For teams already deep in Go services, that's a compelling optimization.
It's still early days though. The repo is actively developed, so expect some rough edges. But if you're building anything agentic that touches sensitive data — especially something latency-sensitive — it's worth keeping an eye on.
Github: github.com/hoophq/alcatraz
One thing I'd love to see more of is multilingual support beyond the current country coverage. Still, for now, this feels like a smart move toward faster, lighter-weight privacy controls without sacrificing accuracy.