Lamplit Light
AI that runs where you are
About
Lamplit Light is Lamplit Labs' sovereign AI lab for small, specialised models you host yourself. Each model is built for one clearly defined job, grounded in verified knowledge, measured against deterministic offline release gates and an independent multi-model judge panel, then compressed into compact ONNX exports for self-hosted use. Every model ships under the light-<purpose>-<version> naming pattern together with its purpose, base model, size, licence and gate result, so you always know what a model is for and which release you are running. The family started with a German language tutor and grows one job at a time.
Key Features
Runs where you are
Host compact models in a browser, on a laptop, on-prem, in a private cloud, inside your app, or on edge and IoT hardware.
Your weights, your rules
Models are designed for self-hosting so your data stays where you run them and the exported weights are yours.
Small by design
Each model is trained for a single, clearly defined job and nothing else, which keeps it small enough to run on hardware you already have.
Gated before release
Every release is measured against deterministic offline gates and checked by an independent multi-model judge panel.
One job per model
Named light-<purpose>-<version> and shipped with its purpose, base model, size, licence and gate result, so you always know what a model is for and which release you run.
Follow Lamplit Light's development
Small, specialised AI models you host yourself in the browser, on a laptop, on-prem, in your private cloud, inside your app, or on the edge. Your data never leaves.