If you don't run your own LLM proxy, how the fuck do you even know whats going on?
Summary
The author argues that to truly understand LLM activity, you should run and instrument your own proxy, rather than rely on provider dashboards. It highlights benefits like multi-provider endpoints, detailed logs, and data control, and shares practical tips for instrumenting traffic and extending the proxy. The piece advocates hands-on visibility and control over data paths and inference workflows.