Automatically route coding tasks to the cheapest Claude model tier that can handle them.
Re-run a task on a bigger model with one command when the cheap tier gets it wrong.
Try a zero-risk audit mode that logs routing decisions without actually delegating.
| scylla23/modelrouter | 100/rutgers-pbl-dining-2015 | a15n/a15n_old | |
|---|---|---|---|
| Stars | 0 | — | — |
| Language | HTML | HTML | HTML |
| Last pushed | — | 2015-12-01 | 2016-06-18 |
| Maintenance | — | Dormant | Dormant |
| Setup difficulty | easy | easy | easy |
| Complexity | 2/5 | 1/5 | 1/5 |
| Audience | developer | general | general |
Figures from each repo's GitHub metadata at analysis time.
Installs as a Claude Code plugin with a single marketplace command, no API keys or config needed.
ModelRouter is a plugin for Claude Code, the command line coding assistant from Anthropic, that automatically sends each task to the cheapest Claude model that can handle it well, rather than always using the most expensive one. The goal is to help people on a fixed usage plan, like Anthropic's Max plan, make their weekly quota last longer without having to think about which model to pick for each request. Every time the router hands a task to a worker model such as Haiku, Sonnet, Opus, or a model called Fable, it prints a short line explaining which model was chosen and why, so the choice is never hidden. If a routed task turns out wrong for its assigned tier, a single command called router redo reruns that same task on a more capable model one level up, and the tool remembers that correction so it makes a better choice next time in similar situations. Over time it builds up a set of personal rules, stored as a plain text file on your own computer, describing which kinds of tasks you have needed escalated in the past. Technically, the plugin works using three built in Claude Code extension points: hooks, skills, and agents. A hook injects routing instructions before each task, another hook checks new delegations against your learned rules and can bump a task to a bigger model if needed, and a skill can temporarily raise the model for a single complex turn before reverting back afterward. There is no proxy server sitting between you and Anthropic, no API key required beyond your normal Claude Code login, and nothing intercepts network traffic. Everything the tool learns and logs stays in plain markdown and JSON files on your own machine, in a folder called dot-router, and nothing is sent elsewhere. Installation is a single command through the Claude Code plugin marketplace, with no extra configuration needed. The project also ships a watch-only audit mode that logs what it would have routed without actually changing anything, useful for trying it out with no risk. It is released under the MIT license.
A Claude Code plugin that routes each task to the cheapest capable Claude model to stretch a usage plan further.
Mainly HTML. The stack also includes Node.js, Claude Code Hooks, Claude Code Skills.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.