Claude Code + Ollama: ~90% cheaper, but who gets credit?

A new GitHub repository shows how to route Claude Code through Ollama, letting developers use open-source models like Gemma, Qwen, and DeepSeek instead of paid Claude API calls. The claimed savings are roughly 90%. This tutorial targets context-heavy terminal tasks including lints, refactors, and batch file operations that chew through monthly token quotas.

This two-engine approach makes sense. Save the expensive Claude model for architecture decisions and tricky bugs. The free local one can grind through repetitive edits. Developers who've watched their Claude Code usage eat through Pro quotas in days will recognize the problem. A 21-slide walkthrough and copy-paste prompt handle most of the configuration.

But Hacker News commenters spotted something the repo doesn't mention. The actual routing software is @musistudio/claude-code-router by musistudio, which appeared on HN nine months before this article. Coherence Daddy's repository fails to credit the original author, and several people called this out. Some suggested the polished presentation is really just advertising for Coherence Daddy's website.

Do you even need an LLM for grep-and-replace? Tools like sed already handle these tasks. Clever setup, but possibly solving a problem that doesn't require LLMs in the first place.