Scenario Builder — chatbot test cases in minutes

Describe your agent in one line and pick the kinds of situations you want to cover. The builder fills fixed scenario templates with your words and gives you up to 10 test scenarios and a checklist. Each category you tick adds up to 3 scenarios, so four or more categories give you the full 10. Copy the result, or download it as Markdown or JSON and keep it in your repo.

This tool runs only in your browser. Nothing you type is sent or stored.

One amber node fanning out into ten branching paths that end in small verdict chips

Build your scenarios

One short line is enough. It is used word for word in the scenarios.

0/200

Examples
What should the scenarios cover?

Each category adds up to 3 scenarios. Tick four or more to get the full 10.

Start with one line about your agent, for example "booking assistant for a dental clinic". Then choose what to cover and select Build scenarios. You'll get up to 10 scenarios.

How to use your scenarios

  1. Download the Markdown file and add it to your repo next to your prompt.
  2. Run each scenario against your agent by hand or in your own test harness. The JSON export is ready for scripts.
  3. Score the conversations with the Transcript Review Scorecard.
  4. Keep the failures as regression tests for the next prompt change.

Questions about Scenario Builder

Does the tool write the scenarios with a language model?

No. It fills fixed templates with the words you type. The same input always gives the same output.

Why did I get fewer than 10 scenarios?

Each category has 3 templates, and the builder never repeats one. One category gives 3 scenarios, two give 6 and three give 9. Tick four or more categories to get the full 10.

Can I edit the scenarios?

Yes. Copy or download them and edit them anywhere. They are yours.

Is my agent description saved?

No. It stays in this browser tab and is gone when you close or clear it.

Does it work for voice agents?

The scenarios are text, so you can use them as scripts for voice tests too. The tool itself doesn't make calls or process audio.

Next step

Want to hear what comes next?

Join early access