I’ve been trying to understand where LLM costs actually come from in real systems, and it’s rarely just “too many tokens in one request”. Most of the time it’s how things behave over multiple calls. Retries keep firing. Flows continue longer than expected. Context grows step by step. Each request looks valid on its own, so nothing stops it — until you notice the cost after the fact. Existing tools helped me see usage, but not act on it in time. So I built Rheonic around a simple idea: put limits and decisions before requests go out, not just visibility after. It can: • enforce request and token limits • detect patterns like retries, loops, and growing context • clamp or block when needed • work across multiple providers Still early, but it’s already helped surface behavior that was completely invisible before. Curious how others are handling this today — are you just watching usage, or do you have something in place to control it while it’s happening?

Rheonic

Limit LLM API usage before it gets expensive

Limit LLM API usage before it gets expensive