Building and Testing Agents

Agents

Agents allow you to build the backend of your generative AI application. You can import an agent template or create your own with your own models and data sets. The Playground allows you to test the output of your agent against different prompts.

Before creating an agent, make sure your project has at least one model available, either added to the project or shared with every project. A data source is optional; add one when you want the agent to answer from your own content. For more information, visit Model Lifecycle Management.

Creating a New Agent

To create a new Agent:

1. Select the Project you want to create your Agent in.
2. Select the Agents tab.
3. Check the Models tab to confirm the project has at least one model.
4. Click Add a New Agent. The agent opens in the builder with Input, a model step, and Output already connected.
5. Click the agent name in the top left to rename it.
6. Select the model step to choose the model and prompt, and drag any other steps you need onto the canvas.
7. Use the line drawer to connect the steps together.
8. Your changes save automatically to your draft (the top bar shows Saving... and then Saved). Click Publish when you want this version to go live for every interface.

Test your agent in Agent Studio

You can now test your agent directly inside Agent Studio.
No more switching back and forth to a separate Playground while you’re iterating on your flow.

Inline Test mode (inside Agent Studio)

Inline Test mode lets you build, test, and refine in one place.

How it works

1. Open your agent in Agent Studio.
2. Open the Playground panel.
3. Type a message (for example, a realistic user input) and send it.
4. Watch how your agent runs through the steps on the canvas and returns a response.

The Playground runs your current draft, so every change you make is immediately testable.

When to use it

Use Inline Test mode when you want to:

• Quickly validate a single flow or prompt change.
• Tweak steps, retry, and iterate in a tight loop.
• Stay in a focused build environment without switching screens.

Compare versions side by side

Use the Compare Playground when you want to see how a change affects responses.

1. Open your agent in Agent Studio.
2. Click the Open Compare Playground icon, just below the Open Playground icon on the canvas.
3. The Test page opens with two panes: your current draft and the agent's published version. You can change the agent and version in each pane, and use + and - next to Agents to show up to five panes.
4. Enter a message once. Every pane runs the same input so you can compare the responses.
5. Click Exit to return to the builder.

Editing an Agent

To edit an Agent:

1. Open the agent.
2. Make changes on the canvas. They save automatically to your draft.
3. Click Publish when you want the changes to go live. Until then, every interface keeps running the published version.

For drafts, published versions, and restoring an earlier version, see Manage Agent Versions.

Debugging an Agent

To debug an Agent:

1. Test the Agent in the Playground.
2. In the left menu, go to Audit → Activity → Agent Executions and select the execution you are debugging.
3. Review the logs for errors.

Using Your Agent

Published agents are used through interfaces. Open the ⋯ menu next to the agent in the Agents list and select Interfaces, or open Interfaces from the builder:

• Airia Catalog lets people in your organization chat with the agent after they sign in to Airia. Under Permissions, choose Public (anyone in your account) or Private and select the users and groups who should see it. See Airia Catalog Interface.
• API is available for every agent, for calling it from your own application. See API Interface.

To Generate Your API Key

You can generate an API key from Settings → Developer → API Keys.

You name the key, set its scope, and copy the key value before you leave the page. The full key is only shown once.

For complete interface setup guidance, see Managing Agent Interfaces.