Most AI teams lose momentum in setup, not in model quality. The goal of this workflow is to move from checkout to your first successful response in minutes, with no shell access and no server hardening work.
1) Provision the stack
After checkout, provisioning starts automatically. Keep the page open and wait until the instance status is marked running. Do not copy gateway credentials before provisioning is complete; they can rotate during initialization.
2) Add one provider key
You only need one model provider key to start: OpenAI, Google, or Anthropic. Add exactly one valid key first, test the agent, and add fallbacks later if needed.
3) Configure channel and pair
If you use Telegram, paste a valid BotFather token. Then open the gateway URL and complete pairing. If pairing is pending, run openclaw devices list and approve the request.
4) Run health + first prompt
Use a simple smoke test:
- Ask the bot a one-line prompt.
- Confirm response latency.
- Confirm logs show a clean request lifecycle.
Why this is faster than VPS setup
- No VM bootstrap scripts to maintain.
- No reverse proxy or TLS termination debugging.
- No manual firewall and origin policy drift.
- Built-in operational defaults for restart and health checks.
Production checklist
- Store gateway token in a secret manager.
- Restrict allowed origins for control UI.
- Set alerting on restarts and health failures.
- Rotate provider/channel credentials on schedule.
If you manage customer instances, this repeatable flow reduces support tickets and improves first-day activation.