Private by design
No prompts sent to a cloud. Sensitive notes, drafts, and data stay on the machine you are sitting at.
BUILT · OPEN SOURCE · SIDE PROJECT
Pocket LLM: A Private AI That Fits in Your Pocket.
An 8-billion-parameter AI chat that runs straight from a USB thumb drive, on Mac, Windows, or Linux. No internet. No cloud account. No subscription. Plug it in, click start, and chat. Nothing you type ever leaves the machine.
Model
8B parameters
On the drive
~5.3 GB
Internet needed
None
GPU needed
None
SEE IT RUNNING
Real Screenshots. Real Answers.
Captured on a 2019 Intel laptop with no GPU. Every reply below was generated locally, on the machine.
WHY I BUILT IT
AI You Own, Not AI You Rent.
Works offline
On a plane, in a secure facility, or anywhere the Wi-Fi gives up. Once it is set up, it needs nothing else.
Portable
One drive, three operating systems. Your settings and chat history travel with the drive.
No subscription
Open-source model, open-source runtime. The only cost is the thumb drive.
It is the same thinking I bring to AI inside revenue organizations: understand what runs where, keep control of your data, and do not lock yourself into a single vendor.
HOW IT WORKS
Three Files and a Click.
01
The runtime
Mozilla's llamafile: a single program that runs the model and serves a chat page, on any OS.
02
The model
Dolphin 2.9.4 on Llama 3.1 8B, quantized to about 4.9 GB so it runs on an everyday laptop CPU.
03
The launchers
My one-click start and stop scripts for Mac, Windows, and Linux. They wait for the model, open the chat, and shut down cleanly.
The GitHub repo has the scripts plus a step-by-step guide: which files to download, how to pick a model size for your hardware, and fixes for the common snags.
Build your ownABOUT THE BUILDER
By Day, I Build Revenue Engines.
I am James Hahn. I lead Revenue Operations and go-to-market for B2B technology companies, from early stage through IPO. I build things like this because hands-on work keeps my AI strategy grounded in what actually runs.