Detects and redacts personally identifiable information (PII), credentials, and sensitive content in AI agent traffic. Built on OpenAI’s open-source privacy filter, quantized to int8 and packaged for fast, fully-local inference with ONNX Runtime.
| Task | PII & sensitive-content detection, redaction |
| Base model | OpenAI privacy filter (open source) |
| Quantization | int8 |
| Runtime | ONNX Runtime, CPU — fully local |
| Catches | Names, emails, phones, addresses, API keys, tokens, card numbers |
| License | See model card |
Download the weights from Hugging Face and run them locally with ONNX Runtime — a minimal Python example is included in the model card. Everything runs on your machine; nothing is sent to a server.
Get it on Hugging FaceThe Privacy Filter is the first in a family of guard models we’re releasing for agentic safety. Follow us on Hugging Face to catch the next releases.