This is OpenAI's 1.5B parameter token classifier for detecting and masking PII in text. It catches eight categories including names, emails, phone numbers, addresses, dates of birth, account numbers, credentials, and personal URLs. The CLI and Python API both support single-pass redaction with a 128k context window, and you can finetune it on your own labeled JSONL if the default model misses domain-specific identifiers. The operating point toggle is handy: high recall casts a wider net with more false positives, high precision is stricter. Runs on CPU or GPU, Apache 2.0 licensed, and the Viterbi decoding keeps span boundaries cleaner than naive token taggers.
npx skills add https://github.com/aradotso/trending-skills --skill openai-privacy-filter