SteadIO Blog

SteadIO is an open-source cost control plane for LLM APIs. This is where we write about what we learn while building it.

Real experiments. Real numbers. Honest post-mortems.

We deliberately ran a runaway agent loop overnight to measure what would happen before anyone noticed. Six hours, $1,847, zero alerts from the provider. That is the kind of thing we write about.

What we cover

LLM cost attribution, proxy architecture, agent reliability, supply chain risk, and the actual economics of running AI in production. We are not writing about AI trends or prompt engineering. We are writing about infrastructure.

Posts worth starting with:

• Why Your LLM Bill Is a Black Box -- the cost attribution problem, explained
• LLM Proxy vs. SDK Wrappers -- why we built a proxy instead of wrapping the SDK
• After the LiteLLM Supply Chain Attack -- what actually changed in 2025
• We Simulated a Runaway Agent and Left It Running Overnight -- exactly what it sounds like

Who writes this

Jonathan Hutchins, founder of SteadIO. I started the company after watching a retry loop generate a $47,000 invoice at a previous job. The invoice arrived two days after the incident. No attribution, no alert, and no way to trace it without reading raw logs.

I am building the tool I wish had existed then.

The source code is on GitHub. Early access to the hosted product is at steadio.ai.

Subscribe

Get new posts when they ship. No digest newsletters.