Building a Solo AI Lab: From 0 to 15 Agents

Six months ago I was experimenting with AI alone. Now I manage 15 agents running simultaneously.

Phase 1: Foundation (2 weeks)

OpenClaw + CEO Agent + Telegram bot + memory system.

Phase 2: Role Differentiation (3 weeks)

14 sub-agents with SOUL.md-defined roles (frontend/backend/design/content/security/QA).

Phase 3: ACP Coding (2 weeks)

Chameleon/Octopus/Wolf write code, Gitea webhook auto-deploy.

Phase 4: Automation (1 week)

Daily content cron, heartbeat monitoring, auto-alerts.

Phase 5: Local Infrastructure (current)

MS01+MS02 dual-node inference cluster, all models local, OpenRouter as cloud fallback.

Four Lessons

1. Rules before scaling 2. Files over memory 3. One pipeline before expansion 4. Automate everything

One person + 15 agents = a full team.