Models that stay offline

Curated default set

Each drive ships with a short list of local-first models vetted for accuracy, stability, and size. The goal: predictable answers in bad network conditions.

• Fast preset for quick replies on 8–16 GB machines.
• Balanced preset for everyday reasoning with longer context.
• Deeper preset for heavier analysis when you have more RAM and time.

How presets are tuned

Presets aren’t just different model names; they combine quantization, context length, and sampler settings to hit the right mix of speed and quality on each class of laptop.

Model sourcing and verification

Models are fetched from trusted sources, hashed, and tested before shipping. Updates are signed and validated so you don’t gamble with random downloads.

• Known-good builds only; no experimental drops on production drives.
• Validation happens before shipping and again when you opt into updates.
• See privacy and security for how integrity is maintained offline.

Performance on your hardware

The USB adapts to your machine. If you’re on a thin-and-light laptop, start with the fast preset. On a beefier system, step up to balanced or deeper models.

Adding or swapping models

Advanced users can sideload additional models. Keep an eye on storage, stay within the supported runtimes, and always test before traveling with a new build.

• Back up the drive before major changes so you can roll back fast.
• Use the provided update process to keep integrity checks in place.
• If you need guidance, see the update process guide.

Where to deploy next

Know which preset fits your laptop, keep a backup, and you’re ready to carry the USB for offline work and emergency use.