Presenter notes: Track conversations, model performance, spans, costs, latency, and traces in LLM applications – all as regular PostHog events - roughly 10x cheaper than other LLM observability tools.