If you don't run your own LLM proxy, how do you know whats going on?

A Hacker News discussion argues that users relying on hosted LLM frontends get little useful visibility into token usage, since provider analytics panels offer limited data for engineers. The author says running a self-hosted LLM proxy gives control over logs (which key called which model, from which client, when, and with what status), virtual API keys, and the ability to switch models and providers quickly. The author reports modifying their own proxy for better insight and custom MCP endpoints, and says they can diagnose a cache miss in about 20 seconds by hand.

Why it matters

Observability and cost control over LLM usage affect how teams debug, budget, and switch providers. The post frames a self-hosted proxy as a practical way to keep that visibility.

Read with care

Source claims not independently checked; the post is an opinion from one user's experience, not a benchmark or survey.