How I cut my OpenClaw token costs 75% after the Anthropic billing change

[devonakelley]

A few weeks ago I noticed my Anthropic bill was growing faster than it should. I run OpenClaw continuously and every task was hitting Sonnet. Scraping a page, summarizing a doc, classifying a record. Same model, same rate, regardless of what the work actually was.

I added a routing layer that does semantic analysis on each incoming request and sends it to the cheapest model that will still perform well on that type of work. Low-complexity tasks go to DeepSeek or Llama, which are 10-50x cheaper. Sonnet, Opus, or Claude Code only fires when the task actually needs it.

The difference from OpenRouter: OpenRouter lets you pick a model. This analyzes what the work is and picks for you, then runs evals on the output so it isn't just blindly routing to cheap and hoping. You get the savings without getting garbage back.

Setup is two commands and a paste:

pip install kalibr
kalibr prompt --openclaw --email you@example.com

Paste the output into your agent. It creates your account, installs dependencies, configures the plugin, and verifies. Zero human steps after that. Free to start at dashboard.kalibr.systems

Happy to answer questions on how it works.