Models and sizes FAQ

Which model for 8 GB RAM?

Use 4–8B quantized models with the low-spec preset and concise prompts.

When to use a larger model?

For long-form, nuanced reasoning on hardware with 16 GB+ RAM and solid ports.

Can I swap models per task?

Yes. Keep a manifest and switch presets based on task complexity and hardware.

Do bigger models drain battery faster?

Yes. Run them plugged in; use smaller models on battery.

How to keep small models sharp?

Give clear context, cap tokens, and use bullet outputs for precision.

Is GPU required?

No. PortableMind is tuned for CPU; modern CPUs handle the included models.

Can I remove models to save space?

Yes, but keep backups. Ensure at least one low-spec and one standard model remain.

How to add a new model safely?

Download on a trusted network, copy to the models folder, and validate offline.

Do models auto-update?

No. Update manually when you choose, then validate offline performance.

Outputs feel generic

Add examples, use stricter formatting, or step up one model size if hardware allows.

Responses are slow

Drop to a smaller model, shorten prompts, and reduce stream speed.

How to avoid rambling?

Set token caps, request bullet points, and include tone/style constraints.