Olyx - One integration. Every AI request governed and traced.
by•
Most teams hardcode a provider SDK, ship fast, and immediately lose visibility into what ran, what it actually cost, or whether sensitive data leaked.
Olyx sits in the request path — one base URL change. Policy enforcement, PII redaction, cost-aware model routing, and a tamper-proof audit trail on every request. No stack rewrite. Provider credentials stay in your environment.
Built for engineering teams past the prototype.
Replies
Olyx started as an internal tool inside an AI infrastructure platform I was building. I was routing between a frontier model for complex data extrapolation and a private model for routine tasks—but I had no real-time way to verify if a routing change was actually saving money, or if it had quietly started leaking PII to an unapproved endpoint.
Cost tracking was a bolted-on afterthought. Auditing production meant manually stitching together scattered provider and application logs. I spent more time maintaining proxy infrastructure than building actual product features.
So I scrapped the multi-language library approach — maintaining the same logic across Ruby, Python, and TypeScript codebases is a nightmare. We went CLI-first instead. Your stack points at one endpoint. The language it's written in stops mattering.
Policy enforcement and PII masking happen before the call reaches the provider. Every routing decision is recorded from request one — tamper-proof and exportable to meet EU AI Act record-keeping requirements.
Early access is open — link in the post. Happy to answer anything on the routing engine, architecture, or what you're running into in your own stack.