Launching tomorrow: An OS for AI agents (solo build, 2,936 tests, 20-page paper)

Hey PH community,

Tomorrow I'm launching Qualixar OS — a runtime that handles the 80% of multi-agent AI work that isn't building agents: routing, quality control, cost tracking, memory, team design.

Quick context on what this is:

I was running 4 AI agents across 3 frameworks. Every day was chaos — no shared memory, no cost visibility, no quality checks. I realized agents need what programs got decades ago: an operating system.

So I built one. Solo, after work hours.

Then in March, a catastrophic rm -rf deleted everything. I rebuilt it from architecture docs. The second version came out cleaner.

What ships tomorrow:

- Forge AI: describe your goal, it designs the agent team automatically

- 13 execution topologies with formal semantics

- Judge pipeline with multi-judge consensus (the quality layer most agent systems skip)

- 24-tab dashboard, 25 MCP tools, 9 built-in tools

- Local-first: SQLite memory, Ollama support, no cloud dependency

- 2,936 tests and a 20-page published paper (arXiv: 2604.06392)

Before shipping, I ran 7 independent AI agents to audit the codebase. They found 76 issues. Fixed all 76.

I'd love to hear from this community:

What's the one operational headache you hit most when running AI agents? Routing? Cost? Quality? Memory? Something else?

That feedback shapes what I prioritize after launch.

See you tomorrow.