From template to live in 15 minutes: what actually happens when you deploy a Claw

Pick a template

Every template's detail page shows the full agent roster before you commit to anything: how many agents, what each one does, which tools the template needs, and an estimated monthly AI cost, so there's no guessing what you're signing up for. You're browsing by category too, money-making, creator, business ops, vertical, lifestyle, not scrolling one long undifferentiated list. Start from an official template, a community one, or a blank canvas if you'd rather build the team from scratch.

[Screenshot: template detail page, agent roster, required tools, estimated cost]

Two answers before anything deploys

Two things stand between you and deploying:

• An API key for whichever model provider your template's agents run on. It's validated live before you move on, and encrypted the moment it's saved.

• Your tools. Anything the template requires gets called out clearly, and you connect it by pasting a key or clicking through an OAuth popup. Don't like a default? Swap it for an equivalent in the same category, HubSpot for Pipedrive, and nothing else about the team changes. If a tool isn't in the catalog at all, you wire up a custom API connection instead, without waiting on anyone to add it for you.

[Screenshot: setup wizard, tool connection step]

Deploy

One button. The container comes up, health checks run against it, and if any step fails, you get the exact step, the actual error, and a retry option, not a generic timeout with nothing to act on.

[Screenshot: dashboard, deployment complete, all health checks green]

Then you connect where it'll actually talk to you

Web chat is already on, nothing to configure. Beyond that, it's your choice: WhatsApp is a QR scan, Telegram is a bot token from BotFather, Slack is one OAuth click. Connect at least one, and a short confirmation screen hands you straight into the live dashboard.

[Screenshot: connect channels step, WhatsApp QR code and Slack OAuth]

The first thing you'll see is a rejection

The dashboard doesn't make you dig for anything. Four numbers sit at the top the moment it loads: agents running, tasks completed today, today's API spend, and how many things are sitting in pending review. Everything else scrolls underneath, newest first, agent by agent.

On the very first cycle, the Orchestrator delegates a task, a Worker produces something, and it goes to the Critic before it ever reaches you. Watch the feed long enough and you'll usually catch a rejection: a score below the threshold, sent back to the Worker with specific feedback instead of a vague "try again." That's the review working exactly as designed. Once something does clear the bar, it doesn't just appear in your inbox, it lands in a pending approvals list with Approve, Reject, and Review sitting right next to it. You decide what actually goes out, every time, not just on day one.

[Screenshot: activity feed, Critic rejecting a task, score and feedback visible]

None of the fifteen minutes above required a terminal, an engineer, or a weekend lost to Dockerfiles. Every piece of setup that would otherwise be yours to own indefinitely, plus everything that comes after launch day, security patches, channel maintenance, monitoring you'd have to wire up yourself, gets compressed into one guided flow and handed back to you as a button.