What is subagent-model-routing?

subagent-model-routing is a tool for people who pay for more than one AI coding assistant, such as Claude Code, Codex, Grok, Kimi, or GLM, and want to delegate a task from one to another without manually copying a prompt into a second terminal and pasting the answer back. It adds a set of small command line shims, one for each supported model family, that send a prompt to another AI coding tool, wait for it to finish, and hand the result back with a clear success or failure signal. Alongside the shims, the project includes a routing skill that decides when a task should be delegated and which model is the right fit for it, plus guardrails that catch cases where a delegated task silently fails, such as missing output or an unexpected error code, so a broken response never gets mistaken for a working one. It also keeps a ledger recording how long each delegation took, whether it succeeded, and other details, so future routing decisions can be based on what has actually worked well before rather than guesswork. The project explains that it is not the same as an API request router like the tools that pick which backend answers a single completion request. Instead, it hands off entire chunks of agentic work, complete with file editing and command execution, to a full separate coding assistant, and then checks that the work genuinely finished. It is meant to sit alongside tools like OpenCode rather than replace them, adding verification and a decision-making layer on top. Installing it requires Python 3.11 or newer, the GNU timeout command, and already having the other AI coding CLIs you want to route to installed and logged in separately. Setup is handled through a bootstrap installer script or a manual clone-and-install process, with separate installation steps for Claude Code, Codex, and other supported clients. The full readme is longer than what was shown.