Contents

  1. 1AI News & ModelsWhich model, what changed, what it costs.
  2. 2Agents & AI CodingWhat works, what breaks, how to run them safely.
  3. 3Local AI LabPrivate AI on hardware you own.
  4. 4AI Security & SovereigntyKeep your data and your agents on a leash.
  5. 5Off-Grid AI & CommsMesh, radio, solar, and offline knowledge.
Subscribe to the Signal Report

Tools

VRAM fit calculator

A quick estimate of how much memory a local model needs. It is arithmetic, not a measurement.

The formula

weights (GB) ≈ parameters (billions) × bits per weight ÷ 8

estimate (GB) ≈ weights × headroom

The headroom covers the context window and runtime overhead. Longer context needs more. Real use varies by runtime, model design, and settings, so treat the result as a starting point and check the model’s own page for its file size.

Reading the result

  • Fits: the estimate is under about 90% of your memory.
  • Tight: it may run, but close other apps and expect short context.
  • Does not fit: pick a smaller model or a lower-bit version.

On a Mac, memory is shared with the system, so leave several GB for macOS.