Running your own LLM proxy for observability
The gist
Developer Jared Siegel says engineers running LLMs in production should build their own proxy layer.
Provider dashboards lack the routing, caching and cost data engineers need to act on.
How it unfolded
- date unknownSiegel publishes an article arguing provider analytics panels are "a distraction" and that engineers should run their own LLM proxy somewhere they control.
- Oct 19, 2026LavX News publishes a report that software engineers running LLMs in production are building their own proxy layers, citing insufficient observability and control over model routing, caching and costs.
- Oct 19, 2026daily.dev posts a summary of the piece advocating a self-hosted LLM proxy for multi-provider, multi-model routing through a single endpoint and detailed logs of which key called what.