Hey. If you’re just getting into AI, this one’s for you.

You’ve probably noticed something weird: the “smart” cloud models are great… and also kind of pricey if you use them for everything. Saying “thanks” or “what’s for dinner?” doesn’t need the same brain as “help me rebuild this whole app.”

That’s where a model router comes in.

What’s a model router, in normal words?

Imagine a helpful friend who sits between you and a bunch of AIs.

You keep chatting in your usual app. Behind the scenes, that friend peeks at what you asked and picks a sensible option:

  • Easy stuff → a model on your computer (often free after setup)
  • Hard stuff → a stronger cloud model (the one you pay for by how much text you send)

You don’t have to keep opening settings and switching models. You just talk.

Trash Panda’s version of that friend is a small free tool that runs on your PC. Open source is here: github.com/jkhall316/trashpanda-router

Why bother?

Cloud AI usually charges by tokens—little bits of text going in and out. Big models cost more per bit.

If every tiny chat hits a premium cloud model, you’re paying sports-car prices for short trips to the store.

A router helps you:

  • Save money on casual chat
  • Still get power when the job is real work
  • Stay in one place—same chat window, less fiddling

It’s less “new AI product” and more “smart traffic director.”

How do I use it? (high level)

  1. Install Docker Desktop (it runs little apps in boxes—don’t overthink it).
  2. Get the Trash Panda router from GitHub and start it (the README walks you through it).
  3. Open the simple admin page in your browser and set a couple basics (like how much graphics memory you have, and cloud keys if you want cloud).
  4. In your chat app, point it at the router once and pick the model called trashpanda/auto.
  5. Chat normally.

That’s it. Your keys stay on your machine. Trash Panda’s website only helps with which model looks like a good fit—it doesn’t take over your accounts.

Works with common chat apps and coding helpers that can use a custom “OpenAI-style” connection (Open WebUI, OpenCode, and friends). Details live in the GitHub README—this page is just the why and the vibe.

Don’t want to install anything yet?

Totally fine. Use the clicky tools on this site first:

When you’re ready for “just chat and let it choose,” come back to the router.

One more thing

You don’t need to become an engineer to use AI well. You need habits that don’t waste money or attention.

A model router is one of those habits: easy questions stay cheap, hard questions can cost more—on purpose.

Friendly next step: skim the router on GitHub, or poke around Pick a model if you’re still exploring.