PIIGhost: a Python library for PII anonymization in LLM agents

I've been building agents on top of LangGraph for a while now, and I keep running into the same problem: every message sent to the LLM might contain sensitive data, and depending on the provider you're using, what happens to that data changes completely. To simplify, there are three families of providers: Non-EU cloud (OpenAI, Anthropic, Google): the best models, but data leaves the EU, which is problematic on many fronts. I wrote a summary here . Sovereign EU cloud (Mistral, Aleph Alpha): proce