Personal OS uses a Chief of Staff architecture — one coordinating agent that owns your daily rhythm and delegates specialized work to purpose-built agents. You never have to think about which agent to use. The Chief of Staff routes automatically.
The Chief of Staff Pattern
When you open a session, the Chief of Staff (Helm) is who you’re talking to. Helm reads your dashboard, understands your current context, and handles the full daily loop — morning reflection, triage, planning, weekly review. Most days, you interact only with Helm.
The key insight is that not all work is the same. Morning journaling requires emotional intelligence and continuity. Breaking down a vague initiative requires strategic depth. Writing code requires precision and execution speed. Reviewing a plan before it ships requires skepticism. Using a single agent for all of this means compromises everywhere.
Instead, when Helm encounters work that a specialist handles better, it delegates. You see the handoff, but you don’t have to manage it.
When Delegation Happens
The Chief of Staff delegates proactively in predictable situations:
- Vague goal or big initiative → strategic-planner breaks it down into phases, milestones, and next actions
- Technology choice or tool comparison → research-scout gathers facts before a recommendation is made
- Defined task ready for implementation → executor builds it cleanly and completely
- Significant work ready to ship → ruthless-reviewer finds problems before they reach the real world
- Long output or complex summary → executive-briefer cuts it to what actually matters
- Calendar overload or prioritization → calendar-priority-mgr analyzes commitments and designs the day
- Decision made or lesson learned → memory-keeper documents it so future you has context
The pattern for complex work is always: Plan → Execute → Review. Atlas structures the work. Mason builds it. Flint checks it.
Model Tiers
Not every task needs the most powerful model. Personal OS assigns agents to model tiers based on what their work requires:
| Tier | Model | Used For |
|---|---|---|
| Reasoning | Claude Opus | Deep analysis, strategic planning, quality review — tasks where errors are expensive |
| Workhorse | Claude Sonnet | Implementation, research, synthesis — the majority of daily work |
| Speed | Claude Haiku | Quick lookups, memory operations, brief formatting tasks |
Strategic-planner and ruthless-reviewer always run on Opus — their outputs directly influence important decisions and the cost of a mistake is high. Executor and research-scout run on Sonnet, where capability meets efficiency. Memory-keeper and executive-briefer run on Haiku because their operations are well-defined and speed matters.
Specialist Modules
Beyond the core team, Personal OS supports specialist agents that activate with optional modules: health data analysis, structured learning, relationship management, content and brand strategy, financial tracking, and career coaching. These agents plug into the same Chief of Staff loop and become active when you run module-specific commands like /health, /study, or /network.
This means the system grows with you. Start with the core team. Add modules as your needs evolve.