Your AI has a dial you have never touched. It quietly decides whether a task comes back as work you would send, or work you would delete. Stop overpaying for the simple jobs, and stop underpowering the hard ones.
Here is exactly what it is. The Model Router is a small kit you drop into Claude Code in about ten seconds. It adds four ready-made helpers and one simple rules file, so your assistant automatically uses the right AI model and the right amount of effort for every task, with nothing for you to choose. You get better work, a lower bill, and your private work stays on your own machine.
Installs with one line in your terminal. Right now, access is by early invite code.
Picture two mornings. In the first, you ask your AI for a warm note to your daughter, and it comes back grammatically perfect and completely dead. In the second, you ask it to check a formula and it thinks for three slow minutes about a three second job.
Same tool, both times. What changed the result was not your prompt. It was two settings underneath it that almost nobody adjusts: which brain does the work, and how hard it thinks.
The best writer and the best planner are not the same brain. Once you know that, you stop asking one setting to do every job.
You ask for something heartfelt and it comes back polished and lifeless.
The Writer uses the strongest language model, so it sounds like a person, like you.
A quick question turns into a slow wait while it overthinks.
Simple jobs run at low effort, so you get a fast answer instead of a spinner.
You are paying premium prices for sorting and tidying.
Busywork goes to the cheap fast option, so you stop paying top dollar for grunt work.
You hesitate to put client or private work into a cloud AI.
Anything private stays on a model on your own machine. It never leaves the building.
You never know which model or setting to choose, so you leave it on default.
You never have to choose again. It picks the right one for every task, on its own.
Your hardest problems get a shallow, hurried answer.
Real problems get the deep brain and more effort, so the answer actually holds up.
Some models are fast and cheap, some are deep, and one is a far better writer than the rest. Match the brain to the kind of work, not to how important it feels. The Model Router sends each job to the right one, on its own.
Effort is how long it reasons before it answers. Cranking it to the top on everything is just slower and dearer, and can make a simple answer worse. The rule turns it up as the task gets harder, and no sooner.
Each helper is a small file that lives in your Claude Code setup, already pointed at the right model and the right effort. You just say which one, or let your assistant choose for you.
Messages, copy, posts, names, stories. It uses the strongest writing model, so the words actually sound human, not like a chatbot.
Big jobs, large refactors, and deep research that would run for a while. It keeps going and checks its own work as it goes.
Careful thinking, decisions, and a second pair of eyes on the work that matters. The one you send anything you cannot afford to get wrong.
Sorting, tagging, tidying, rough first drafts at scale. The cheap fast option, so you never pay premium prices for busywork.
Imagine opening your AI tomorrow morning and every answer already arrives at the right depth, in the right voice, on the right brain, without you touching a single setting. That is the week you are about to have.
Your writing stops sounding like a chatbot, because the right model writes it. Less rewriting, more that lands the first time.
You stop running everything at the most expensive setting, so the bill drops while the quality goes up where it counts.
You never pick a model or an effort level again. The rule chooses for you, quietly, on every single task.
Client records and sensitive work never touch a cloud model. Sovereignty is built into the very first rule.
Quick jobs get quick answers, so you stop watching a spinner for something that needed three seconds.
Real problems get the deep brain and more effort, so the answers hold up when it matters most.
This is built for solo founders, creators, coaches, and small business owners who use Claude Code in their terminal or editor, and who are not hardcore engineers.
If you want your private work to stay private, and you want to stop running every prompt at the most expensive setting, this is for you.
It is not for anyone who only uses the Claude chat app in a browser. It needs the Claude Code environment.
Enter your early access code, and land on your own walkthrough page straight away.
Download the kit and run one line in your terminal.
Restart Claude Code, and it starts routing on its own.
Works with Claude Code in your terminal, the VS Code or JetBrains extension, or the Claude desktop app Code tab. Not a chat setting. No second brain needed.
Mac or Linux install with one line. Windows works too, through WSL or a quick by-hand copy.
The Model Router is in early access right now, so you need an invite code to get in. If you were invited, use the code from your invite. If not, the free lesson below explains the whole idea, and more codes open up regularly.
Yes. It names model tiers, not exact products, so it keeps working as models change, and it falls back to your strong default if you do not have the top creative model yet.
It never sends anything on its own. It only decides which brain thinks about a task, and your privacy rule keeps sensitive work on your own machine.
Yes. The install is one line and the walkthrough holds your hand, with by-hand steps too. If you can copy and paste, you can do this.
Claude Code in your terminal, the VS Code or JetBrains extension, or the Claude desktop app Code tab. It is not a chat setting, so it does not apply to the Claude chat app or Claude in a browser.
One payment. Yours to keep. A founding price, with free updates as the models change.
Right now, access is by invite code. Enter yours below, or sign in if you already unlocked it.
Prefer to try the idea first? The free lesson explains the whole thing, with nothing to install.