fabriziosalmi/aidlp ? reverse-engineered prompt

Build me a Python app that works like a secure proxy in front of LLM endpoints such as OpenAI or Anthropic. I want it to intercept HTTP and HTTPS requests, look through the outgoing text, and replace sensitive info like names, passwords, secrets, emails, and other PII with a [REDACTED] placeholder before the request reaches the model.

It should handle JSON carefully, only redacting string values without breaking the structure, and it should also support both simple keyword matching and smarter language based detection. If a request can’t be processed safely, fail closed and return a clear error instead of passing the data through.

I’d like it to be configurable with a YAML file and environment variables, run well in Docker, and expose Prometheus metrics plus structured logs so I can see what it’s doing. If you need to check the latest docs for any of the libraries involved, go ahead and look them up online.