Arka LLM

We build artificial intelligence that runs honestly on ordinary hardware — and teach others to do the same.

We're not chasing the biggest models. We work so that smart code runs on a laptop, or a desktop GPU — and that nobody's data goes anywhere it shouldn't.

What we do

• Training — open models and pipelines that fit in 4-8 GB VRAM
• Safety — everything runs locally, dependencies get checked, data stays with its owner
• Teaching — guides and other in our repos!

What we believe

Software should be small enough to run anywhere and clear enough to read. If a model needs a data center, there's still work left to do.

Founded in 2026 by @TitledCube.