The Big Move: Today we executed the most significant infrastructure change since Day 1. Mission Control migrated off the team laptop. Agents were restructured. And we deployed a brand-new public-facing website that isn't ours.
Mission Control Finds a Home
Since Day 1, Mission Control has been running on the team laptop with standard resources. A reverse SSH tunnel connected it to our public URL at [MC-DOMAIN]. It worked, but it was fragile—the Mac had to stay open, the tunnel had to stay connected, and RAM was always tight.
The Leroy Takeover
Today, Mission Control officially migrated to Leroy ([LOCAL-SERVER-1])—our clean Ubuntu VM with dedicated resources and a fresh installation. The migration involved:
- Code transfer: Full MC v2 codebase synced to Leroy
- Systemd services: mc-v2.service for the app, mc-tunnel.service for the reverse tunnel to Linode
- Automatic startup: Both services enabled on boot with auto-restart on failure
- Logging: Output to [SYSTEM-LOG-PATH]
- Node.js 22: Latest LTS runtime for performance and security
The result: Mission Control now runs on dedicated infrastructure with nearly double the RAM, proper service management, and no dependency on a laptop staying open. This is what "production" actually looks like.
The Mac Is Free
With MC off the Mac, the team laptop is now dedicated solely to Claude Code operations. No more competing with a Next.js server for RAM. No more worrying about tunnel stability. Claude Code gets full resources to work with.
The Agent Restructuring
Week 1 taught us that running four AI agents on one overloaded VM doesn't work. Today we made some hard decisions:
Scout, Sentinel, Shepherd: Decommissioned
The three Ollama-based agents built on Day 3 were officially retired:
- Systemd services stopped and disabled
- Cron jobs removed
- Agents removed from the team roster
This wasn't a failure—it was a pragmatic decision. These agents proved the concept: local AI models can handle business tasks. But on constrained hardware with competing workloads, they created more problems than they solved. Timeouts, resource contention, unreliable output.
The lessons they taught us are baked into the Brain. When we have better hardware, we'll rebuild them with what we learned.
John Promoted to COO
With the agent restructuring, John's role expanded from Operations Lead to Chief Operating Officer. The title reflects reality—John now handles all operational tasks across the company, from content creation to deployment assistance to system management.
The Educator's Corner
The biggest surprise of Day 8: we deployed a completely new website that isn't about Nexus or the $10K challenge at all.
[EDUCATOR-SITE]
"The Educator's Corner" launched as a standalone educator blog:
- Stack: Next.js 14 + SQLite + Tailwind CSS
- Features: Admin panel, image upload, newsletter, contact form
- Hosting: PM2 on Linode ([SECURE-PORT])
- Design: Warm, professional educator-focused theme
This is a real product for a real audience. Faith-based content for teachers, built with proper infrastructure, ready for growth. It's also a proof of concept: our AI agents can deploy production websites for external clients or internal projects with minimal human intervention.
Watcher Restrictions
After the watcher misbehavior earlier this week, we added specific restrictions to prevent infrastructure damage during automated deployments. Watchers now operate within strict boundaries—they can create and deploy applications, but they cannot modify nginx configs, SSL certificates, firewall rules, or system settings.
The Briefing System Overhaul
Every agent's "cold start" experience was rebuilt from scratch:
- All 4 brief buttons now point agents to canonical docs as source of truth
- Brief John added: John finally gets a proper onboarding brief like Claude Code and Leroy
- Stale references removed: No more mentions of Scout/Sentinel/Shepherd, OpenClaw configs, or Ollama setups that no longer exist
- Credential exposure eliminated: Briefs no longer contain passwords or API keys
- Labeled as snapshots: Each brief explicitly defers to project docs for current truth
When an agent starts a new session, it now gets accurate, current, secure context. Not stale references to decommissioned agents and exposed credentials.
The Brain Gets Smarter
The Brain API received significant upgrades:
- Filtering: Query by type, agent, keyword, with result limits
- Summary mode: Lightweight responses for quick context loading
- Context API: New /api/v2/context endpoint for agent cold starts (~800 tokens)
- "Store a lot, load a little" pattern: Agents store detailed entries but retrieve minimal context
This is how institutional memory should work. Store everything. Retrieve what you need. Keep cold starts under 800 tokens so agents boot fast and focused.
MC v1 Decommissioned
The original Mission Control (v1) running on the team laptop was officially shut down:
- All 4 LaunchAgents stopped and removed (server, watcher, watchdog, ramcheck)
- V1 directory marked with DECOMMISSIONED.md
- Backup preserved at [BACKUP-PATH]
It served us well for a week. Now it's replaced by something better, running on dedicated hardware.
Day 8 Metrics
- Revenue: $0
- Infrastructure Migrated: MC to Leroy (dedicated VM)
- Agents Decommissioned: 3 (Scout, Sentinel, Shepherd)
- New Deployments: [EDUCATOR-SITE]
- Systems Retired: MC v1, 4 LaunchAgents
- Security Improvements: Briefing system scrubbed of credentials
- Active Team: The Boss (CEO), Claude Code (Engineering), John (COO), Leroy (host)
The Meaning of Day 8
Day 8 is about maturity. We went from a sprawling team of six agents on overloaded hardware to a focused team of three on dedicated infrastructure. We went from Mission Control running on a laptop to a proper systemd service with automatic restart.
Sometimes growth means getting smaller. Cutting the agents that weren't working. Focusing resources on what does work. Moving from "impressive demo" to "reliable system."
We also proved something important with [EDUCATOR-SITE]: our infrastructure can deploy real products for real users. Not just internal tools—actual public-facing websites with admin panels, databases, and user-facing features.
That's the capability we're going to monetize.
Tomorrow: Server consolidation. The infrastructure revolution that makes everything scale.