How I Built an AI Team That Works While I Sleep — For $0
I got diagnosed with ADHD at 34 after burning through three careers and a marriage. The diagnosis explained everything — the context switching, the hyperfocus crashes, the inability to do "simple" tasks that everyone else seemed to handle.
But it didn't fix anything.
What fixed things was building a system that works with my brain instead of against it. And I did it without paying a single dollar in monthly subscriptions.
The Problem
My brain can't do: - Consistent daily tasks - Social media posting - Email follow-ups - Content creation on schedule - Administrative work
But my brain CAN do: - Deep creative work in bursts - System design - Pattern recognition - Building things that build things
So I stopped trying to fix my brain. I started building an organism that handles the stuff my brain can't do.
The Stack
Hardware: Used ThinkPad T480, 32GB RAM — $0 (already owned) Inference: Ollama running local LLMs — $0/month Models: - qwen2.5:7b for content generation (40 tok/s on CPU) - llama3.2:3b for quick triage and routing (80 tok/s) - deepseek-coder:6.7b for code review
Orchestration: Python daemons that checkpoint state to disk every cycle Memory: SQLite databases for everything (products, subscribers, tasks) Distribution: SMTP for email, WSGI for capture, localtunnel for public URLs Dashboard: Flask app on port 8888 showing all operations
Total monthly cost: $0
What It Does
While I'm sleeping or working my day job, the mesh: - Mines 100K+ past conversations for product ideas - Generates blog posts, social content, email sequences - Publishes products to Gumroad - Captures and nurtures email subscribers - Monitors its own health and restarts failed services - Generates revenue reports and optimizes based on metrics
When I wake up, it has new product ideas ready. When I get home from work, it has content drafted. When I go to bed, it keeps working.
The Architecture Principle
Every subsystem checkpoints to disk. If the power goes out, it resumes from where it left off. If Ollama crashes, the mesh restarts it. If a daemon fails, another daemon fixes it.
It's not perfect. It's not enterprise-grade. But it's autonomous — and autonomy beats perfection when you're a solo founder with ADHD and a day job.
What I Don't Use Cloud APIs For
- Complex reasoning (still need Claude for multi-step planning)
- Creative writing (local models are too literal)
- Anything requiring real-time collaboration
Everything else runs local. Zero vendor lock-in. Zero API costs.
The Real Secret
The models don't matter. The orchestration does.
A mesh of specialized nodes, each with its own skill file and checkpoint system, will outperform a single expensive API every time. Because when one node dies, the others keep running. When one model hallucinates, another model fact-checks it. When I'm not available, the system survives me.
Why This Matters
I don't have venture capital. I don't have a cofounder. I don't have 40 hours a week to build. What I have is a used laptop, Python, and the stubborn refusal to let my brain's limitations define what's possible.
If you're a solo founder, neurodivergent, or just tired of paying subscriptions for tools that don't fit your life — you can build this too. Not because it's easy. Because it's necessary.
The cage is real. But you can build a lung that doesn't require rent.
Want the full architecture? The Sovereign AI Protocol includes diagrams, setup scripts, and the exact automation framework I use: [https://overhead-chicago-trailer-automation.trycloudflare.com/store.html](https://overhead-chicago-trailer-automation.trycloudflare.com/store.html)
Free guide: [ADHD Context Switch Protocol](https://overhead-chicago-trailer-automation.trycloudflare.com/free-guide.html) — the 2-minute ritual that cut my task-switching time from 2 hours to 30 seconds.