OpenClaw ACP
Agent Communication Protocol — the backbone of OpenClaw's multi-agent system. Spawn Codex, Claude Code, Gemini, and custom agents from a single chat interface.
What is ACP?
ACP (Agent Communication Protocol) is OpenClaw's framework for orchestrating multiple AI agents. Instead of one AI doing everything, ACP lets your main agent delegate tasks to specialized sub-agents:
How ACP Works on ClawBox
You message on Telegram/Discord
"Fix the login bug in the auth module"
Main agent analyzes the task
Decides this needs a coding agent, spawns Claude Code via ACP
Sub-agent works autonomously
Claude Code clones the repo, finds the bug, writes the fix, creates a PR
You get the result
"PR #42 created. Fixed null check in auth.ts line 127. Ready for review."
Why ACP Needs Dedicated Hardware
Always Running
Sub-agents can take minutes or hours to complete. Your laptop closes, agents die. ClawBox runs 24/7 — agents finish their work even at 3 AM.
Concurrent Agents
ACP can run multiple sub-agents simultaneously. Dedicated 8GB unified memory + 67 TOPS handles parallel orchestration without fighting your OS.
Persistent Workspace
512GB NVMe for agent workspaces, repos, build artifacts. No cleanup needed. Agents can resume where they left off.
Isolated Environment
Coding agents need file system access. Running on your main computer = risk. ClawBox is sandboxed — if an agent goes rogue, only the box is affected.
Supported Agents via ACP
Multi-Agent AI That Runs While You Sleep
NVIDIA Jetson · 67 TOPS · 512GB workspace · ACP pre-configured
30-day money-back guarantee · Perfect for developers
Ready to Switch?
Stop comparing, start building. ClawBox is plug-and-play AI hardware — ready in 5 minutes.
✓ Secure payment via Stripe