We have been working on an open source routing mechanism that has built in guardrails (i.e. preventing jailbreak, injection, PII leakage). It's really similar to OpenRouter but we've added a guardrail mechanism to it.

These guardrails are highly extensible but have focused around natural language based interactions (e.g. you have a chatbot that you don't want leaking PII).

Some our guardrails are other open source models (like Metas prompt guard they released with 3.1 or Microsoft Presidio).

Would love feedback to see what folks think!

Thanks!