Skip to content

4. Deploy the Agent

What does "deploy" actually mean?

So far, you've only chatted with your agent from your own screen. That's great for testing, but it only works while you're the one testing it.

Deploying means pushing your agent onto real, managed infrastructure — specifically AWS (Amazon's cloud) — so it becomes reachable independently of your own screen. This is what makes it genuinely usable, not just testable.

Think of it like the difference between rehearsing in your living room and performing on a real stage where an audience can actually watch.

Steps

  1. Go to the Agents home page and find your agent's card.

  2. Click "☁ Deploy AWS" on the card.

  3. Choose a target environment from the dropdown.

  4. Typical choices are dev, stage, or prod — exactly which ones you see depends on how your platform install is configured.

  5. Why this matters: different environments let you test safely (dev) before going fully live (prod).

  6. Confirm the deployment.

  7. A deploy progress banner appears on screen.
  8. It shows:
    • A spinner (so you know it's working).
    • A progress bar.
    • Live step messages as the deployment moves along, like "Creating Lambda function..." or "Configuring API Gateway..."

  9. You can watch this banner update in real time until the deployment finishes. No need to refresh the page.

  10. Need to stop it? While a deployment is in progress, click "Cancel Deployment" to stop it.

After deployment finishes

Once your deployment completes, your agent's card shows a status badge:

  • "AWS Deployed" — success! Your agent is live.
  • "AWS Failed" — something went wrong. (Don't panic — this happens, and you can try again.)

From here, you have two more useful buttons:

  • "🔁 Redeploy AWS" — push an updated version of your agent (use this after you make changes).
  • "☁✖ Undeploy AWS" — remove your agent from AWS when you're done with it.

One more thing to know: once deployed, chatting with your agent (as in step 3) automatically uses the AWS-hosted version — even if you never ran anything on your own machine.

✅ Checkpoint: What you just did

Huge milestone — take a moment to appreciate this! 🎉

You now have a complete, deployed agent that:

  • Was created by you, from scratch.
  • Has a real tool it can use.
  • Runs on real cloud infrastructure, independent of your own screen.

That's the entire lifecycle: create → add a tool → test → deploy. You've done it!

What's next

From here, you have two paths:

  • Curious how developers talk to the platform via code instead of clicking? Check out 5. MCP SDK Reference. This is completely optional — read it later, whenever you're ready to go deeper.
  • Want to keep exploring the UI? Look around your agent's page as it matures beyond this "hello world" example. You'll find things like logs, rollback, and scheduling — useful tools for managing a real production agent.