wtf is basicallymythos?

BasicallyMythos is a small local tool that watches Claude Code, the command line coding assistant, for a specific hidden behavior: when a safety safeguard trips on a request, Claude Code quietly re-runs that request on a different, more expensive model without telling you. This project lets you see every one of those silent switches, and lets you pick which model catches the fallback instead of accepting whatever Claude Code chooses by default. The tool itself is a tiny proxy server, about 100 lines of code with no external dependencies, that runs on your own computer. Normally, all your Claude Code traffic flows straight through to Anthropic's servers unchanged. But when a request gets flagged and rerouted, the proxy catches it and sends that specific request to Kimi K3, an open-weights model with a large context window, using your own separate Kimi subscription key. Every reroute gets logged with a timestamp so the downgrade is finally visible instead of invisible. Setting it up means copying the proxy file into your Claude Code configuration folder, storing your Kimi API key securely in the Mac keychain rather than in a plain text file, setting up a background service so the proxy keeps running, and merging a small settings snippet into your Claude Code configuration. The project documents several specific gotchas it ran into, such as Kimi subscription keys needing a different web address than regular Moonshot platform keys, and a setting that must be turned on or the reroute never triggers at all. Everything runs only on your own machine, and your normal Anthropic login and subscription pass through completely untouched. The Kimi key is only ever used for the specific requests that get rerouted. The project is currently macOS only, with support for Linux and Windows planned. It is open source under the MIT license, and the author is active answering questions and taking feedback on social media.