LLM logs are crushing my application logging system. We recently launched AI features on our app and went from ~100mb/month of normal website logs to 3gb/month of llm conversation logs and growing. Our existing logging system was overwhelmed (queries timing out, etc), and costs started increasing. We’re considering how to re-architect our llm logs specifically so we can handle more users plus the increasing token use from things like reasoning models, tool calling, and multi-agent systems. I’m not selling any solutions here, genuinely curious what others are doing. Do you store them alongside APM logs? Dedicated LLM logging service? Build it yourself with open source tools?