Most AI tools give you one model's answer. LetThemBuild gives you a room: several models work on the same question together, one leads, the others review it, and you get a single agreed answer with a record of how it was reached.
On the Mac, Windows and Linux app the room has hands. It works in a real folder on your machine — reading files, writing them, running your build and your tests — so "it works" is something a reviewer can check rather than something the lead asserts.
The two apps
LetThemBuild is the desktop app, the command line and the editor extensions. It is where the work happens, because it is the only place with your checkout.
LetThemChat is the web app at letthemchat.com, with the iPhone app on its way. Same account, same discussions, kept in sync. You can read and continue anything from there, and hand work back to your Mac to run.
What makes it different
Four things, in the order they matter.
A review here is evidence, not an opinion. A reviewer does not read the lead's answer and nod. It runs its own commands — your build, your tests, whatever settles the question — and its verdict carries those commands and their exit codes. An answer everyone agreed with, where nobody opened a file, is labelled as exactly that.
You get a receipt. Every finished discussion shows what was run, how much passed, what changed and what was found. It is the difference between "the AI says it fixed it" and knowing which test proved it.
Disagreement is the product. One model's confident mistake is the failure case of every single-model tool. Here another model has to sign it off, and when it will not, you see the argument rather than the average.
You are not locked to one lab. Claude, ChatGPT, Gemini and Grok out of the box, any OpenRouter model in any seat, your own endpoint if you have one, and the vendors' own CLIs if you already pay for them. Bring your keys or use ours.
What it is not
It is not a chatbot with a bigger context window, and it is not an autonomous agent you point at a repository and leave for a week. It is a working session you watch, interrupt and steer, where the checking is done by something other than the thing that wrote the code.