The model picker that picks itself.

flux@store: ~/run

$ cursor-auto-switcher --run

✓ point it at your input

✓ run the command

✓ it processes locally

› get structured output

done — output ready ▍

How it works

Point it at your input

Run the command

It processes locally

Get structured output

Why it beats the alternative

This product

A paid SaaS tool

Monthly cost

$0 — runs on your machine

$80–150/mo, billed forever

Setup

Drop in & run in minutes

Accounts, API keys, rate limits

Your data

Never leaves your computer

Uploaded to their cloud

Ownership

Yours forever — edit the code

Rented; stops when you cancel

How it actually works

Exactly what it does — and when.

1Quick build / refactor → routes to the FAST agentic coder (Cursor composer) so simple edits ship in seconds, not minutes.

2Gnarly bug — concurrency, a deadlock, a hairy algorithm → escalates to the MAX-EFFORT coder (gpt-5-codex, high reasoning) that actually solves it.

3Plan / reason / long-context read → routes to the FAST GENERALIST (grok, 1M context) instead of burning a coder model on prose.

4One config line turns it on; it logs every routing decision + the reason, so you can see why each task went where.

In practice · A normal coding session

You ask for a small UI tweak — it routes to the fast agentic coder and the diff lands in seconds. Ten minutes later you hit a race condition that's been haunting you; the router sees the difficulty and escalates to the max-effort coder, which traces it and fixes it. Then you ask it to plan a refactor across 40 files — it switches to the long-context generalist to hold the whole codebase at once. You never picked a model. It picked right, every time, and logged why.

Use it to…

Auto-route build tasks

Feature and boilerplate work is detected and sent to the fast agentic coder so you ship builds quicker.

Escalate hard problems

Tricky logic and gnarly debugging route to the max-effort coder automatically — no manual model swap.

Offload reasoning work

General reasoning and analysis go to the fast generalist, keeping latency low without you thinking about it.

The shift it creates

Without it

$80–150/mo SaaS bills stacking up
Rate caps + vendor lock-in
Data you don't actually own

→

With it

Stop hand-picking a model for every task
Know exactly why each model was chosen
Go live in under a minute

What's inside

Per-task real-time model routing

One config line to turn it on

The full drop-in router logic (~15 lines)

Works with Cursor or any multi-model CLI/IDE

Transparent — logs every choice + why

Instant download — no email, no payment

Questions, answered

✓ Free · no email✓ Yours to own✓ Use on every project✓ Built to ship results