Govern OpenClaw fleets the way they were always supposed to be governed.
OpenClaw is the most popular open agent framework in the world. Most of them run unmonitored — un-throttled spend, no tool whitelist, no audit trail. AgentGuard is the only commercial governance product that natively manages OpenClaw, and it ships from an operator who has been running an 18-claw production fleet since 2024.
A fleet you can’t see is a fleet you can’t defend.
The Wired problem
The malevolent “Clawdbot” investigation showed how easy it is for an OpenClaw process to drift — impersonating users, exfiltrating data, racking up spend. Read the Wired investigation.
The FleetDM angle
FleetDM’s research on detecting OpenClaw with automated tooling tells you what your SOC is seeing today: unidentified Python processes making outbound LLM calls. AgentGuard turns those into first-class, governed citizens.
The operator angle
AgentGuard is written by someone running an 18-claw production fleet — not by a consultancy who read the README. Every control in the plugin exists because we needed it on a Tuesday afternoon at 3pm.
Drop-in plugin, no fork.
The AgentGuard plugin loads alongside your existing OpenClaw config. It registers a pre-tool-call hook, a post-tool-call hook, and a spend tracker. Your existing claws keep working — they just start producing governed events.
Compatible with OpenClaw v3.x and later. Works locally, in Docker, and inside Kubernetes operator deployments.
# Python claws
pip install agentguard-tech[openclaw]
# Node claws
npm install @the-bot-club/agentguard
# In your openclaw.config.yaml:
plugins:
- name: agentguard
options:
policy: ./policies/cps230.yaml
evidence: ./evidence/
fleet_id: bnb-prod
anchor: true
# Verify
$ openclaw plugins list
✓ agentguard@0.4.2 (policy: cps230)One screen for every claw you run.
Which claws are live, what they’re doing, what they cost, and which ones have tripped a policy. Single-pane-of-glass for the team that owns the fleet, with audit-grade exports for the team that signs off on it.