Show HN: blackbar-nano – local PII detection and redaction on CPU (github.com)

A new tool called blackbar-nano has been unveiled, designed for local detection and redaction of personally identifiable information (PII) without the need for GPU acceleration. This solution can identify and redact 35 types of PII, processing messages in approximately 38 ms on an M1 CPU. Unlike many existing systems that rely on remote API calls, blackbar-nano operates entirely on the user's machine, ensuring that sensitive text never leaves the local environment—an essential feature for maintaining privacy in data handling. This development is significant for the AI/ML community as it addresses the growing demand for secure, local-first solutions in PII management, especially critical for industries handling sensitive data such as healthcare and finance. The model is built using a variant of ModernBERT-base and has shown impressive metrics: achieving a redaction coverage of 96.9% while maintaining a lower false-alarm rate compared to similar tools like GLiNER2-PII. The tradeoff mechanisms for entity and label confidence enable users to customize the threshold for detection based on their specific needs, enhancing flexibility and control in real-time applications.