What you leave with
- A Console organisation with separate keys for development and production
- An OpenClaw assistant connected to a messaging channel of your choice
- Knowledge grounded in your own documents, with answers citing their source
- Daily usage and cost reporting, with a hard ceiling
- An escalation path to a human when the assistant is unsure
Day one — foundations
| Block | Content |
|---|
| 09.00–10.30 | Console organisation, workspaces, keys, usage limits |
| 10.45–12.00 | Model choice for an assistant: quality against cost per conversation |
| 13.00–14.30 | Install OpenClaw, connect Claude, first reply |
| 14.45–16.00 | Prompt structure: persona, boundaries, refusal behaviour |
Day two — knowledge and channels
| Block | Content |
|---|
| 09.00–10.30 | Feeding your documents: chunking, retrieval, keeping answers grounded |
| 10.45–12.00 | Connecting WhatsApp, Telegram or a web widget |
| 13.00–14.30 | Conversation state, handover, and what to remember |
| 14.45–16.00 | Practice: the assistant answers twenty real questions |
Day three — production
| Block | Content |
|---|
| 09.00–10.30 | Deploying on Rumus Cloud: environment, secrets, releases |
| 10.45–12.00 | Monitoring quality: sample review, escalation rate, unanswered share |
| 13.00–14.30 | Cost control: caching, model routing, spend ceiling |
| 14.45–16.00 | Launch checklist and rollout to your team |
Channel requirement. Decide before day two which channel the assistant will
serve. Preparing the account takes longer than the session allows, and Telegram or the web
widget are the quickest to arrange.