Coding models can produce good code, but a plausible diff is not the same as finished software.

Agent OS is an open-source, local-first AI Project Operating System that builds the surrounding harness:

• persistent project memory

• Main Agent planning and orchestration

• sandboxed Coding Agent execution

• real build, browser, and visual verification

• evidence-backed, bounded recovery

• explicit approval before Git, deployment, database, or payment-related actions

To pressure-test the system, I used Agent OS itself to build, repair, verify, and deploy Pulseboard, a full-stack SaaS, from an empty repository.

The 1-min demo uses those real run records. There is no staged interface, fabricated state, or LLM re-run.

This is the first public release, built independently as a university student. Windows currently has the most thoroughly tested setup, while macOS and Linux use manual installation.

I’d especially appreciate feedback on:

1. Whether the product is understandable from the demo

2. Whether the brain/hands separation makes sense

3. Where the onboarding or installation feels too complicated