NEW: Runtime Safety for AI Agents

Your OpenClaw Agent Has Zero Guardrails

Add network allow-listing, human approval gates, rate limiting, and multi-scanner skill vetting — in under 60 seconds.

Free tier available · 1 Guard instance · 5 vetted skills · No credit card required

Install in 60 seconds

docker run -d --name custos-guard \
-e CUSTOS_API_KEY=your_key_here \
-p 8080:8080 \
vettiq/guard:latest

Then point your OpenClaw instance to http://localhost:8080 as the proxy.

Six Layers of Runtime Protection

Goes beyond what ClawHub's built-in scanning catches

Network Allow-listing

Block all outbound traffic except vetted skill APIs. Zero-trust by default.

Human Approval Gates

Require approval for sensitive operations: email sends, file deletes, external POSTs.

Multi-Scanner Skill Vetting

4-scanner pipeline: Snyk, Cisco, Semgrep, VirusTotal. Detect malicious skills before they run.

Live Action Log

Real-time visibility into every skill call, API request, and approval decision.

Rate Limiting

Prevent runaway agents. Set action limits per minute, hour, or day.

Fail-Closed Architecture

If Guard crashes, traffic stops. Nothing leaks.

Three Policy Profiles

Choose the right security posture for your use case

Profile
Sandbox
Evaluation / demos
Team
Daily team work
Production
Customer-facing workflows
NetworkBlock all except skill APIsSkill APIs + common SaaSSkill APIs + strict logging
EmailBlockedDrafts OK, sending needs approvalAllowed, bulk needs approval
File SystemRead-only sandboxRead-write in workspaceRead-write, approval outside workspace
Approval GatesNone (risky = blocked)Email, file deletes, ext. POSTsBulk, financial, destructive ops
Rate Limit10 actions/min30 actions/min60 actions/min
SkillsLow-risk onlyLow + medium riskAll approved skills

15 Vetted Skills at Launch

Scanned by Snyk, Cisco, Semgrep, and VirusTotal

Email/Messaging

4 skills

Gmail, Slack, SMTP, Outlook

Calendar

3 skills

Google Calendar, Outlook Calendar, Cal.com

Documents

4 skills

Google Docs, Notion, Local File Manager, Web Scraper

DevOps

4 skills

GitHub, Shell Runner, Docker Manager, PostgreSQL

Can't find a skill you need? Request vetting at contact@vettiq.ai

Frequently Asked Questions

How does Guard work without modifying OpenClaw?

Guard runs as an HTTP proxy. OpenClaw's outbound traffic routes through it. No plugins, no patches, no OpenClaw modifications.

What happens if Guard goes down?

Fail-closed architecture. If the proxy crashes, OpenClaw's traffic has nowhere to go. Nothing leaks.

Do I need Docker?

Yes. Guard runs as a Docker container alongside OpenClaw.

What skills are in the catalog?

15 vetted skills across 4 categories: Email/Messaging (Gmail, Slack, SMTP, Outlook), Calendar (Google Calendar, Outlook Calendar, Cal.com), Documents (Google Docs, Notion, Local File Manager, Web Scraper), DevOps (GitHub, Shell Runner, Docker Manager, PostgreSQL).

Can I use skills not in the catalog?

Yes, but they won't have vetted allow-lists. Guard logs all their activity and blocks unknown domains. Use Sandbox profile for unvetted skills.

Is my data sent to your servers?

Action logs and heartbeats, yes. Your actual OpenClaw conversations and tool outputs stay local. Guard sees request destinations, not payloads.

Add Guardrails in 60 Seconds

One Docker command. Zero OpenClaw modifications. Full runtime visibility.