---
schema: 1
type: term
title: '**model switcher**'
description: >-
  **model switcher** — an infrastructure piece at Val Town that lets the product
  run multiple AI models at once and absorb model progress by changing a model
  name. Steve Krouse describes it, alongside beta flags, as part of his goal to
  build so the tool keeps working if the underlying LLM stays the same and
  captures the benefits when a better model ships. [^ebfee90b@0]
canonical: >-
  https://liquidgold.cc/channel/everyinc/explore/term/model-switcher-8fa4f986f141
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/model-switcher-8fa4f986f141.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*model switcher\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/model-switcher-8fa4f986f141

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**model switcher** — an infrastructure piece at Val Town that lets the product run multiple AI models at once and absorb model progress by changing a model name. Steve Krouse describes it, alongside beta flags, as part of his goal to build so the tool keeps working if the underlying LLM stays the same and captures the benefits when a better model ships. [^ebfee90b@0]

## What this page helps you discover — **model switcher** in this channel's material ### The context where the term appears In this episode, Val Town's founder Steve Krouse explains what it takes to build a product that stays neutral to AI's rate of progress. His stated goal: if the underlying LLM stays the same, the tool keeps working — and if a better model ships, Val Town gets the benefits by changing a model name. The **model switcher** is one of the infrastructure pieces that make that possible. [^ebfee90b@0] ### What the model switcher does The **model switcher** lets Val Town run multiple models at once, alongside beta flags — so the team can ship, test, and roll back between models without rebuilding the product. [^ebfee90b@0] ### Why it matters in the broader argument Krouse pairs the switcher with a few supporting pieces that show how much engineering model progress demands: - **Beta flags** to run multiple models simultaneously - **Usage-based pricing**, urgently needed once new Townie users started running expensive models - **Per-model prompt iteration** — in practice, Val Town actually rolled back from a newer Sonnet because they couldn't get its prompting right In other words, the switcher isn't just a convenience: it's what lets a product absorb model progress without being hostage to it — the way Cursor and WebSim jumped the moment Sonnet 3.5 arrived. [^ebfee90b@0] ## Key takeaways - A **model switcher** is the infrastructure piece that lets a product run multiple AI models at once and change which one it uses by updating a model name. - It exists so a product can stay **neutral to model progress**: keep working if the LLM stays the same, capture gains if it improves. - Val Town uses it alongside beta flags, usage-based pricing, and per-model prompt iteration — and even with all that, prompting can still fail badly enough to force a rollback. [^ebfee90b@0]