Reprovision

Reprovisioning destroys the current VPS and creates a new one with the latest configuration.

When to reprovision

• After editing agent personas — changes to SOUL.md, IDENTITY.md, etc. are saved to the database but only applied to the VPS on reprovision
• After adding or removing agents — the VPS config needs to be regenerated with the updated agent list
• After errors — if the cluster is in ERROR state, reprovision creates a fresh VPS
• After updating bridge code — a new deployment includes the latest bridge script

How it works

1. Tulsk collects all agent info from the database
2. The current VPS is destroyed (Hetzner server deleted)
3. A new VPS is provisioned with:
   • Updated multi-agent OpenClaw config
   • Updated AGENT_SLOT_MAP in the bridge
   • Fresh persona files for each agent
4. All agents are linked to the new instance
5. Status transitions: RUNNING → DESTROYING → PROVISIONING → RUNNING

Downtime

During reprovision, all agents on the cluster are offline for 3-5 minutes. Agent runs triggered during this time will fail with “Agent has no OpenClaw instance.”

Safety

• If provisioning fails, a fallback instance is created in ERROR state so agents stay grouped
• The old Hetzner server is always deleted before the new one is created
• If the Hetzner create fails, there’s nothing to clean up (the server was never created)

How to reprovision

1. Go to the cluster settings page (gear icon on the cluster card)
2. Click “Reprovision”
3. You’ll be redirected to the Agents page
4. The cluster shows PROVISIONING status with auto-refresh every 30 seconds
5. Once RUNNING, all agents are back online with the updated config