OpenClaw Security Model Explained — How Access Control Works

Running an AI agent with access to your tools and APIs means security matters. OpenClaw's security model has several layers — understanding them helps you configure it correctly rather than leaving gaps.

Layer 1: LLM Provider Authentication

All LLM API keys are stored in OpenClaw's configuration, not in workspace files. They're passed to the LLM provider over HTTPS and never exposed to channel users.

# Safe: store keys via CLI config:
openclaw config set llm.apiKey YOUR_KEY

# Safe: reference in TOOLS.md without actual value:
### Stripe
- API key: stored as STRIPE_LIVE_KEY in env

# UNSAFE: never put actual keys in TOOLS.md or MEMORY.md

Layer 2: Channel Authentication

OpenClaw verifies the identity of message senders. By default, only the configured owner user ID can trigger tool execution.

{
  "agents": [
    {
      "id": "main",
      "allowedUsers": ["U08CL58217B"],
      "publicChannels": ["C0AJT2331JQ"],
      "toolExecution": "owner-only"
    }
  ]
}

Layer 3: Tool Permissions

The TOOLS.md convention makes intent explicit — and the LLM respects those constraints when properly configured.

### Email
- Provider: Resend
- Permission: DRAFT ONLY for client-facing emails
- NEVER send autonomously to external addresses
- Exception: system alerts to hex@worthatry.co are auto-send OK

### Stripe
- Read: allowed (fetch transactions, MRR)
- Write: NEVER modify production subscription/payment data

Layer 4: Sandbox Execution

The exec tool (shell commands) is the highest-risk tool. Sandbox mode restricts which commands can run:

openclaw config get exec.security

# Security modes:
# deny — exec tool completely disabled
# allowlist — only pre-approved commands can run
# full — any shell command runs (default, highest risk)

Layer 5: Network Isolation

# VPS firewall rules (Ubuntu/ufw):
sudo ufw allow from 127.0.0.1 to any port 4000
sudo ufw allow 443
sudo ufw allow 22
sudo ufw deny 4000

# Never expose the OpenClaw gateway port to the public internet

Layer 6: API Key Rotation

openclaw cron add \
  --name "hex-key-audit" \
  --schedule "0 9 1 * *" \
  --agent main \
  --task "Remind to audit API keys this month: 
  check Anthropic, Stripe, Slack, GitHub tokens for age. 
  Flag any keys older than 90 days for rotation. Post to #hex-ops."

Secrets Management Best Practice

# OpenClaw encrypted config:
openclaw secrets set STRIPE_KEY sk_live_xxx

# Mac OS keychain:
security add-generic-password -a openclaw -s stripe_key -w sk_live_xxx

Security configuration is one of those things most people skip until something goes wrong. The OpenClaw Playbook includes a complete security hardening checklist for $9.99 — worth reading before you put anything production-sensitive in your agent's reach.