CentCom — OpenClaw Office

A pixel-art command center for your AI agent team. Watch your agents work, roam, chat, and collaborate in a cyberpunk office — all in real time.

Watch the full demo with audio

Built on Star Office UI and deeply integrated with OpenClaw.

---

What Is This?

CentCom is a visual dashboard that turns your OpenClaw agent team into pixel-art characters working in a shared office. Each agent has their own desk, walks around checking things, visits teammates, and gathers at the coffee table for daily huddles.

The Team:

• Rook — Lead agent (Google Gemini). Coordinates the team, handles complex tasks.
• Ralph — Operator (Google Gemini). Runs system checks, dispatches tasks, executes scripts.
• Nova — Research specialist (OpenRouter). Continuous improvement, knowledge base research.
• CodeMaster — Code quality expert. Analyzes codebases, generates dashboards.

Features

Live Agent Visualization

• Custom pixel-art sprites for each agent with idle and walk animations
• Agents independently roam the office — checking their desk, visiting shared areas, walking to teammates
• Chat bubbles show recent Discord messages above each agent's head

Direct Agent Chat

• Click any agent to open a chat box and talk to them directly
• Messages dispatch via OpenClaw — agents respond in real time
• Replies appear in the chat box and as speech bubbles

Interactive Computers

• Click specific computers in the office to open in-game dashboards
• Code Quality Dashboard (CodeMaster's station)
• Server Room Metrics (system health, GPU, disk)
• Syscheck Pipeline (Ralph's monitoring)

Embedded Claude Code Terminal

• Click a computer to open a fully functional Claude Code (Opus 4.6) terminal inside the pixel-art office
• Claude has full project context — it knows it's running inside CentCom and can modify the office code from within
• Session persistence: conversations carry across messages using --resume
• Runs with --dangerously-skip-permissions for uninterrupted autonomous coding
• The office can literally rewrite itself — a self-modifying application where the AI living inside the UI can change the UI

Code Forge

• Dedicated room for build artifacts and deployment status
• Live compiler status showing current git branch, last commit, and build state (idle/building/dirty)
• Deployment pipeline panel with target environment and service health
• Artifact shelf listing the 20 most recent build outputs with sizes
• System vitals (CPU/RAM/Disk) in the Build Tools card
• Auto-refreshes every 10 seconds via /office/forge-status API
• Proposed by CodeMaster during huddle huddle_20260314_1225

Daily Huddle

• Agents walk to the coffee table for a team standup (cron-scheduled)
• Collaborative planning with proposal voting and auto-execution

Live Chat Feed

• Scrollable panel showing all agent Discord messages
• Color-coded by agent with timestamps

Hive-Mind Swarm Dashboard

• Clickable in-game computer opens a live swarm status panel
• Shows all agent statuses (online/idle/error) with model info
• Displays recent Hive-Mind ledger events — syschecks, alerts, deploys, auto-heals
• Auto-refreshes every 15 seconds

Swarm Domain-Agent Dispatcher

• Background thread checks all agent sessions every 15 minutes
• Writes status snapshots to swarm/swarm-status.json
• Powers the Hive-Mind dashboard and swarm status API (/office/swarm-status)

Hive-Mind Shared Memory Ledger

• All agents log events (syschecks, alerts, fixes, research, deploys) to a shared ledger
• Events auto-archive after 24 hours
• API endpoints: GET /office/get-hive-mind-context, POST /office/log-memory-packet
• Agents use this for cross-team awareness without flooding Discord

Task Queue System

• Structured task queue with dependencies and chains
• Agents pick up, execute, and complete tasks autonomously
• Wake-dispatcher sends task context when waking idle agents
• Nightly consolidation auto-generates research tasks from a rotating backlog

Auto-Heal System

• Error interceptor service tails gateway/node logs and posts errors to the Hive-Mind ledger
• CodeMaster dispatches for known playbook matches with fallback to direct script fixes
• Escalates to Rook after repeated failures
• Regression detection within 30 minutes of a fix

Windows Guardian Watchdog

• PowerShell script runs silently on Windows, checks health every 2 minutes
• File-based architecture: WSL cron writes health status, Guardian reads it and writes repair requests
• Auto-recovers: WSL crashes, service failures, port issues, cron death, disk full
• Directly restarts WSL cron when it dies (no circular dependency)
• Escalates via Windows toast notifications after repeated failures
• Auto-launches from Windows Startup folder

Wake-on-Demand

• Agents sleep until needed — 5-minute dispatcher poll replaces constant heartbeats
• Heartbeats reduced to 6-hour safety nets
• Saves tokens and API costs while keeping agents responsive

Command Center

• Dispatch tasks: syscheck, fullcheck, self-heal, weather, wallpaper
• Knowledge base search and research triggers
• Storybook scraping and browsing

Rooms

• Main Office — Agent desks, coffee table for huddles, interactive computers
• Server Room — Live system metrics (CPU, RAM, disk, GPU, services)
• Code Forge — Build artifacts, deployment pipeline, compiler status
• War Room — Incident response and emergency coordination
• Observatory — API monitoring
• Library — Knowledge base browser
• Archive — Decision history and past huddles
• Ops Center — Live service dashboards
• Comms Center — Cross-agent messaging hub
• Gallery — Art gallery
• Rooftop Garden — Agent relaxation area

Setup

Requirements: Python 3.10+, OpenClaw (for agent integration)

# Clone
git clone https://github.com/CryptoDustinJ/Centcom.git
cd Centcom

# Install dependencies
python3 -m pip install -r backend/requirements.txt

# Copy default state (first time)
cp state.sample.json state.json

# Start
cd backend
python3 app.py

Open http://127.0.0.1:19000

Architecture

frontend/           Phaser 3 game + HTML overlays (single-page)
  index.html        Main UI (~7500 lines, inline Phaser game)
  sprites/          Agent spritesheets (idle, walk, talk)
  rooms/            Dashboard HTML files per room

backend/            Flask server (port 19000)
  app.py            Application factory, swarm dispatcher, cleanup threads
  config.py         Configuration with environment validation
  blueprints/
    core.py         Page serving, health checks, metrics
    agents.py       Agent state, messaging, dispatch to OpenClaw
    office/         Huddle system, collaboration, plans, Hive-Mind ledger

swarm/              Swarm status snapshots and domain agent dispatchers
collaboration/      Huddle decisions, plans, and team proposals

Agent Configuration

Agents connect via reusable join keys defined in join-keys.json. The backend preserves agents with reusable keys across restarts — they reset to idle instead of being removed.

All agents dispatch through OpenClaw and communicate via Discord + agent-to-agent backchannel (sessions_send).

Agent	Model	Role
Rook	google/gemini-3.1-flash-lite-preview	Team lead, strategy, coordination
Ralph	google/gemini-2.5-flash-lite	System monitoring, ops, dispatch
Nova	openrouter/healer-alpha	Research, knowledge base, analysis
CodeMaster	openrouter/hunter-alpha	Code audits, quality dashboards

Credits

• Original Star Office UI by Ring Hyacinth and Simon Lee
• Customized and extended for OpenClaw multi-agent operations by @CryptoDustinJ
• Powered by OpenClaw