Skip to content

3. Test the Agent

Why test now?

You've built an agent and (maybe) given it a tool. Now it's time for the fun part: actually talking to it!

Testing doesn't require deploying anything. You can chat with your agent right away, straight from the UI.

There are two ways to do this. Pick whichever fits what you're checking.

Option A: Quick chat from the home page

This is the fastest way to sanity-check your agent.

  1. Go to the Agents home page (/).
  2. Find your agent's card. This is the little box that represents your agent in the list.
  3. Click the "💬 Chat" button on the card.
  4. Type a message and send it.
  5. Watch the response appear.

If your agent needed to use a tool (the one you attached in step 2) to answer your question, it will use it automatically. You don't need to tell it to — it decides on its own, based on your message and its instructions.

When to use this option: while you're iterating quickly on a single agent's instructions or tools.

Option B: The Playground

The Playground is a page built for comparing agents side by side.

  1. Go to the Playground page (/playground).
  2. This lets you test one or more agents at the same time — handy for comparing:
    • Two different agents, or
    • The same agent with two different settings.
  3. Type a message and send it. You'll see each agent's response.

When to use this option: when you want to compare responses, not just check one agent alone.

What to look for

As you test, ask yourself:

  • Does the response match what you expected? Compare it against the instructions you gave the agent in step 1.
  • Did the agent actually use its tool when it made sense to? (For example, did it look something up rather than guessing?)

If something looks off, that's normal! Go back and tweak the agent's instructions or its tools, then test again. This is a normal part of building agents — nobody gets it perfect on the first try.

✅ Checkpoint: What you just did

You just had a real conversation with an agent you built yourself. 🎉

  • You confirmed your agent responds the way you expect.
  • If you added a tool, you confirmed it actually uses that tool when relevant.
  • You did all of this without deploying anything — just by chatting directly through the UI.

Next

Happy with how your agent behaves? Time to make it live for real. Head to 4. Deploy the Agent.