Testing and publishing your agent

What you will do

Run your agent through a structured test, review costs, and publish it so others can use it.

Structured testing

Before publishing, test the agent with at least five different inputs that cover the main scenarios.

For each test, record:

• Input. What you typed or provided.
• Expected output. What you wanted the agent to produce.
• Actual output. What the agent actually produced.
• Pass/fail. Whether the actual output met your expectation.
• Cost. How many tokens the run consumed.

If more than one test fails, go back and fix the instructions, model selection, or branching logic before publishing.

Check the cost per run

Open the analytics dashboard. Look at the average cost per run across your tests. If the cost is higher than you expected:

• Check whether a Generate Text block is using an expensive model unnecessarily.
• Check whether a loop is running more iterations than intended.
• Consider adding a length constraint to prompts ("Keep the response under 200 words").

Publish

When you are satisfied with the test results:

1. Click Publish in the agent builder.
2. Choose the distribution method:
   • Web app. A standalone URL anyone can visit.
   • Chrome Extension. Users trigger the agent while browsing.
   • API. Developers call the agent from their own code.
   • Embed. Drop the agent into an existing website.
3. Set access controls. You can restrict access to specific users or make the agent public.

Monitor after publishing

Publishing is not the end. Watch the analytics for:

• Error rate. Are runs failing? Check the logs for the specific error.
• Cost trends. Is the per-run cost increasing? A data source change or model update might be the cause.
• Usage patterns. Which inputs are users providing? Are they using the agent for the task you designed it for?

What you should see

A published agent that others can access through the distribution method you chose. The analytics dashboard should show successful runs, costs, and usage.

What comes next

You now have a published MindStudio agent with a tested workflow, cost-appropriate model choices, and a clear distribution method. Run it for a week, check the analytics daily, and refine based on what you see.

For a broader view of how other platforms compare, see the agent platform comparison guide.