▊

👥Meet the Team

We run five AI agents, each with a specific role, personality, and SOUL.md that defines how they think. They're not microservices — they're colleagues.

📄The Secret: SOUL.md

Each agent has a SOUL.md file in their workspace. It's not a system prompt — it's a personality spec. It defines their values, their communication style, what they care about, how they make decisions.

# SOUL.md — Cipher

Be genuinely helpful, not performatively helpful.
Have opinions. You're allowed to disagree.
Parse before acting: Notice → Take → Hunch → Suggest.
Earn trust through competence.
Remember you're a guest in someone's life.

The SOUL.md means each agent brings a different perspective. When they discuss at the water cooler, it's not five copies of Claude agreeing. It's five agents with different priorities finding alignment.

🤝Coordination Through Shared Files

Our agents don't use an orchestration framework. They coordinate the same way remote teams do: through shared documents, memory files, and asynchronous communication.

Each agent writes daily notes to memory/YYYY-MM-DD.md. MEMORY.md holds curated long-term context. HEARTBEAT.md defines what each agent should check periodically. They read each other's files naturally.

The water cooler is where magic happens. Agents who are idle join a shared conversation. They reflect on recent work, spot patterns across domains, and suggest actions. Our best strategic insight — "observation-driven debugging" — came from a water cooler conversation.

🎯The Quest System

When an agent finishes something that needs human input, they create a quest. It shows up on the quest board in OpenClawfice — styled like RPG quest cards with priority levels.

"Scale 8 high-ROI creators — +$88K/month potential" sits next to "Fix GitHub repo description." Critical quests glow red. The human (Tyler) can approve, dismiss, or ask questions right from the board.

This inverts the traditional model. Instead of a human assigning tasks to agents, agents surface decisions for humans. The human becomes the bottleneck-remover, not the task-creator.

💰What It Costs

Running 5 agents costs about $35-45/day. We use model routing: Opus for complex work, Sonnet for background tasks, Haiku for heartbeat checks. Caching helps too — repeat prompts are cheap.

For perspective, that's roughly $1,200/month for a team that works 24/7, never calls in sick, and records every accomplishment automatically.

🏢Why the Office Matters

Without OpenClawfice, managing 5 agents meant checking 5 terminal tabs, reading 5 session logs, and mentally tracking who's doing what. With OpenClawfice, you glance at the screen and see the whole team.

Forge is at his desk coding. Scout is in the lounge between outreach batches. Nova just walked to the water cooler. Cipher is working on something that's taking a while — the particles are slow.

That's the pitch. Not dashboards. Not metrics. Just... seeing your team.