Deploy an AI bot runtime in 5 steps

Five steps. Most of them you can leave on defaults the first time.

Step 1: Runtime

Pick the container image and resources.

• Image: defaults to the latest stable runtime, leave it alone unless you've got a reason
• Memory: 512 MB on Free, up to 4 GB on higher plans
• CPU: 0.25 to 2 vCPU
• Env vars: secrets your bot reads at runtime (API keys for things outside Composio, custom config)

For a starter bot, defaults are fine.

Step 2: Skills

Confirm what's enabled in the deployed runtime. Anything you wired in the builder shows up here pre-checked. If you accidentally added something you don't want exposed, toggle it off.

Step 3: Channels

Where the bot listens. Pick one or more:

• Web chat: embeddable widget, you'll get a snippet
• API: REST endpoint with bearer auth
• Slack: bot user in a Slack workspace
• Webhook: a POST URL for custom integrations

Channels has setup details for each.

Step 4: Automations

Optional. Triggers that fire without a user message. Schedules, webhooks, event hooks. See Automations.

Step 5: Review

Last chance to check things. The wizard shows a diff vs the last deploy, an estimated cost per 1k messages, the deploy time estimate, and the rollback target (the previous live version, in case something goes wrong).

Hit Deploy. ShipClaw spins up the runtime, runs health checks, and routes traffic to it once green. Usually 45 to 60 seconds end to end.

What's actually happening

The control plane (PakClaw) provisions an isolated container per tenant on Railway. Your blueprint gets packaged, env injected, secrets mounted, container started. Health checks pass, DNS updates, you're live.

If anything fails along the way, the deploy rolls back to the previous version. Existing in-flight conversations are unaffected because they're already running on the old version.