yee

Most companies acquire your data to train their models, then rent those models back to you like another SaaS. No shade — we use their software, it's fire. Yee runs the other direction: SFT and RAG tooling are built in, so every code edit you make can become training data for your own LLM, built from open-source models. Bring your own keys, keep it local-first. We just think the intelligence you generate at your keyboard should compound into something you own.

Honest truth: this is niche software, not for the majority. It's for people who enjoy web development — React, Tailwind, HTML, CSS — and think TypeScript is their middle name. It also works great if you live in Python notebooks. If your day is C++ or C#, this isn't for you. And it's not a Claude and it's not a Cursor; if you use those, keep using them — Yee sits next to them as a tamer, more accurate way to edit code. Quick, focused jobs, no marathon runs.

Here's the day-to-day. Your app previews live while you work — Yee stands up the dev server itself and puts the internal dashboard buried in your repo one click away. Things code should do automatically never cost you money here; ask Claude or Cursor to hunt that down and you're paying tokens to start a web server. Real Lighthouse scores sit in the chat header, and anything below 100 is one click from "fix it in chat."

We're hyper-focused on taming frontier models into something potent instead of chasing dragons in caves nobody cares about — and we keep the receipts. We built eval software to measure how under-the-hood prompting raises or lowers a model's token spend, and in our measured runs models spent meaningfully fewer tokens, with far less variance run to run. There's a per-conversation token meter in the app, because we'd rather show you the bill than ask you to trust us about it.

The parts of shipping that usually bite you are covered too. Before you push, Yee fact-checks whether your build will actually deploy — it caught a real frozen-lockfile failure before it left the machine. Nothing is ever lost: a shadow git repo checkpoints every file state, so chats branch, merge, and rebase like git, with a safety net that can recover every original. And when you do push, the pull request writes the story of the session — prompts, tools, files changed — instead of a bare one-liner.

If this goes well, we plan to add a $1 and a $5 plan to unlock the complexities of key generation, for people who want their privacy and their IP handled their own way. That's the whole roadmap we're willing to promise.