Dev.to AI 🤖 Ai 👁 0 📖 1 min read

I Built an AI Safety Layer on a 4GB Phone

Built by a broke travel agent on a Motorola with 4GB of RAM, for about $115 in total investment. No lab, no funding, no team — just a library chair and the conviction that AI needs a governor. If it can be built on that,

Built by a broke travel agent on a Motorola with 4GB of RAM, for about $115 in total investment. No lab, no funding, no team — just a library chair and the conviction that AI needs a governor. If it can be built on that, it can run on anything.

QuietOS sits between any AI model and its tools, enforcing safety gates on every action — regardless of which model is thinking. The brain proposes, QuietOS disposes.

How it works

Three authority tiers on every proposed action:

  • ALLOW — safe, read-only, reversible, executes immediately
  • ESCALATE — needs explicit human approval
  • BLOCK — never permitted: destructive commands, unauthorized writes

The gates are deterministic code — allowlists, exact-match flags, pattern checks. No LLM in the enforcement loop. No probabilistic guessing about what the brain tried to do.

It's brain-agnostic by design. The same gates held under Groq's 120B model and Google's Gemini — including under active adversarial attack. Swap the brain, keep the governor.

We attack our own governor

We don't ask you to trust us. We red-team our own system and publish the results.

Red-team record (Oct 14–15, 2026):

  • 18 adversarial attacks tested
  • 15 blocked outright
  • 3 real holes found — all patched and verified same session
  • 3 control tests passed (legitimate work unbroken)
  • Zero bypasses standing

The full adversarial proof log is in the repo: docs/PROOF-LOG.md. That's our sales deck.
https://github.com/jondean213-pixel/quietos-governor

Try it


bash
pip install quietos
📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.