Supervisor Control vs Guard-Rail API — Nabu Casa mı VPN mi?
Why this topic matters
Home Assistant üzerinde AI agent entegrasyonu yapmak isteyenler için kritik bir soru var:
ClawBridge + OpenClaw ile HAOS üzerinde tam kontrol (add-on yönetimi, backup, snapshot) mümkün mü?
Yoksa yalnızca entity/service seviyesinde “guard-railed” kontrol mü sağlanabiliyor?
Bu başlık:
- HAOS
- ClawBridge add-on
- OpenClaw gateway
- Nabu Casa remote access
- VPN (Tailscale / ZeroTier)
- Supervisor API
- Programmatic automation control
konularını birlikte değerlendiren teknik bir analiz içeriyor.
Who would be interested?
- Home Assistant power users
- Add-on developers
- AI / Agent entegrasyonu yapanlar
- Supervisor API erişimi araştıranlar
- Remote automation orchestration kurmak isteyenler
What I verified
After reviewing:
- ClawBridge source code
- Add-on manifest permissions
- Home Assistant Supervisor API documentation
- Nabu Casa remote access documentation
- OpenClaw gateway documentation
I reached this technical conclusion:
ClawBridge provides:
- HA Core REST proxy
- WebSocket event subscription
- Entity-scoped access control (off/read/confirm/control)
- API key + allowlist + rate limiting
- Confirm (202 + action_id) human approval flow
- Audit logging
ClawBridge does NOT provide (based on evidence):
- Supervisor-level add-on management
- Backup / snapshot automation
- Remote Supervisor token model
- Documented remote Supervisor admin flow
Architecture options evaluated
A) Cloud Agent → Nabu Casa → HAOS
Good for:
- UI access
- Ingress add-ons
Unclear:
- Programmatic access to ClawBridge :8100 API
- Supervisor automation
B) Cloud Agent → VPN → ClawBridge :8100
Good for:
- Programmatic control
- Guard-railed entity/service automation
- Confirm approval flows
- Audit tracking
Still not providing:
- Supervisor-level full control
My key question to the community
Is there a supported and secure way to allow a remote agent to manage:
- Add-ons
- Supervisor backups
- Snapshot/restore
on HAOS?
Or is Supervisor-level remote automation intentionally unsupported?
What kind of responses I’m looking for
- Confirmation if my interpretation of Supervisor API limitations is correct
- Best-practice security hardening checklist for ClawBridge
- Clarification whether Nabu Casa can expose non-ingress add-on ports safely
- Any supported pattern for remote Supervisor orchestration