What it is
OpenClaw is an autonomous agent runtime for business workflows. It receives triggers (GitHub events, webhooks, scheduled tasks), runs multi-step agent pipelines to decide what to do, validates every output through Portarium, and surfaces human-approval requests before executing irreversible actions.
It is the operational layer of the agentic infra stack — where Portarium handles governance, OpenClaw handles execution.
What it runs
Currently used as the operator layer for governed workflow experiments on calvinkennedy.com:
- Inquiry routing — classifies inbound contact form submissions and routes them to the right response template
- Content audit — scans writing and project pages for stale or inconsistent information
- GitHub event processing — responds to repo activity with appropriate documentation or visibility updates
Architecture
Trigger (webhook/schedule)
→ OpenClaw runtime
→ Agent pipeline (Claude)
→ Portarium validation
→ Approved: execute action
→ Needs review: surface to human
→ Human approves/rejects
→ Continue or halt
→ Audit log (every step)
Why self-hosted
The point of governed AI workflows is that you own the policy. SaaS agent platforms own it for you. OpenClaw is being packaged for a low-cost AWS VPS runtime so the execution path stays under direct control, the audit trail remains local, and the stack can be evolved without vendor lock-in.