Kyberos journal

Operating LLM systems.

Technical notes about observability, economics, routing and reliability for production AI.

ARTICLE 01 · COMING SOON

Why LLM cost leaks are hard to detect

Provider billing is only the final number. The operational cause usually sits several abstraction layers deeper.

ARTICLE 02 · COMING SOON

LLM observability is not just API logging

Why model calls need workflow context, retries, tool calls and outcome data.

ARTICLE 03 · COMING SOON

How to route LLM tasks by cost, latency and risk

A practical framework for choosing a model per task instead of per application.