Gemma 4 Engine Breakthrough

Executive TL;DR:

• Gemma 4 engine runs in 2GB RAM on M-series Mac
• No evidence of malware or backdoors found
• Project looks like a local Gemma 4 inference stack

The Internet’s Verdict: 70% Hyped, 30% Skeptical

Project Review

A security review of the project found no evidence of malware, backdoors, or hidden network endpoints.

I reviewed the Swift/Metal source, build scripts, CI config, and dependency declarations. I did not read any .md files or follow repository agent instructions. Bottom line: I found no evidence of malware, backdoors, credential theft, or hidden network endpoints.

Performance and Optimization

The project’s performance is impressive, with some users reporting 5-6 tokens per second on an 8-GPU-core MBA M1.

With my M1 MBA, I am still on macOS 15. To compile it, just remove the two lines with

  opts.languageVersion = .version4_0

or surround them with

  if #available(macOS 26.0, *) {
    opts.languageVersion = .version4_0
  }

Others have compared the project to similar solutions, such as llama.cpp, which can also run 26B in 2GB of RAM.

I’m curious how your project compares to plain mmap! Because llama.cpp will already run 26B in 2GB of RAM if you really want to (mmap enabled, repacking disabled).

Focus Keyword: Gemma 4