AI research, turned into things you'd actually keep using.

We're a small software lab working on applied AI. Our products aren't demos of what AI can do; they're built to be genuinely convenient, so using them stops feeling like "trying out AI" and starts feeling like just getting something done.

Meet Trunk, our first product Our philosophy
12:34
Home
i SETTINGS
This Device
10.9 GB RAM
Suggested model size
up to ~8.3B parameters (Q4)
Library
3 models downloaded

Three things we hold ourselves to

"AI-powered" gets attached to almost anything now. We'd rather ship less and have each thing we ship actually earn a permanent place on your device.

01

Useful, not novel

A feature has to survive past the first week of curiosity. If it's only interesting because it's AI, it's not interesting enough.

02

Private by default

Where it's technically possible, we'd rather run the model on your hardware than build a reason for you to trust us with your data.

03

Research-led, not trend-led

We follow what's actually improving (on-device inference, smaller capable models) over whatever's loudest that quarter.

What we've built so far

Where this comes from

Trunk exists because on-device inference (via llama.cpp and its ecosystem) reached the point where a phone can run a genuinely capable model without a server in the loop. We track that frontier and build when it's actually ready for daily use, not before.

2023
llama.cpp makes CPU inference fast enough to matter outside a research lab.
2024
Quantization shrinks genuinely capable models into a phone's memory budget.
2026
Trunk ships: on-device inference finally convenient enough for daily use.