Quenga Designs · 2026
Quengtopia
A desktop app for running and evolving a language model entirely on your own machine — no API keys, no cloud, nothing leaving the device.
- Problem
- Every practical way to work with a language model assumes a round trip to somebody else's server. That means an API key, a per-token bill, and your prompts — and whatever you paste into them — leaving your machine. For anything personal, that is the wrong default.
- Solution
- A desktop application rather than a web app: a Tauri shell (Rust) around a Vite + React + TypeScript interface, sitting on top of a Python engine that handles evolution, evaluation and checkpointing. The model, the corpus and every checkpoint live on local disk. Nothing is uploaded, and it works with the network off.
- Result
- A working local-first product that runs offline on a normal desktop. Because it is a desktop app by design it has no public URL — this page is the only place to see it, which is the honest trade for keeping the data on the machine.
Local by construction
Quengtopia is a desktop application, not a site with a login. The Tauri shell gives it a native window and direct filesystem access, so the model weights, the training corpus and every checkpoint stay on local disk. There is no API key to configure, no per-token cost, and no request that leaves the machine — it works with the network switched off.
The two halves
The interface is Vite + React + TypeScript, rendered inside Tauri's webview. Underneath, a Python engine does the work that actually takes time: evolving candidates, evaluating them against each other, and writing checkpoints so a run can be stopped and resumed rather than restarted.
Keeping those halves separate matters. The Python side can be long-running and interrupted without taking the interface down with it, and the interface can be rebuilt without touching the engine.
Why there is no demo link
The corpus and accumulated checkpoints run to roughly fourteen gigabytes. Even setting aside the privacy argument, that is not something to host — and hosting it would undo the point of the project. For viewing it from elsewhere on my own network there is a small web server bridged over Tailscale, which is a convenience for me rather than a public deployment.
Where it stands
In active development. It runs, it evolves models, and it is used — but it is a workbench rather than a finished product, which is why it sits under In dev rather than Live.