Add your system and team
You describe your system once, then connect it. Litmus uses your description and the risks you list to suggest the criteria, personas and test cases in the steps that follow.
Before you start
- Have the details in what you need ready.
- Use a screen at least 1,024 pixels wide. Litmus's steps don't fit on smaller screens.
Sign in
- Choose Continue with TechPass. MOE staff without TechPass may choose Continue with Google with their Google Workspace account instead.
- The first time, Litmus asks whether you're new to AI testing. If you are, The basics explains the ideas in five short cards. You can open them again later from your user menu, under Learn the basics.
- To practise first, choose Guided setup on the home page. It walks you through a fictional example system, and nothing you do there is saved.
Describe your system
Choose Add new system. On the first page, fill in:
| Field | What to enter |
|---|---|
| System name | The name your team uses for the system. |
| Agency | The agency that owns the system. |
| Project users | The teammates who will work on the system. See add your team. |
| What it does | What the system does, for whom, and what it must not do. Litmus can rewrite your description for you. |
| Use case type | The kinds of task the system does, such as answering questions, extracting information or classifying text. Choose all that apply. This shapes the criteria Litmus recommends and the personas it suggests. |
| Who uses it | Public officers, teachers, members of the public or students. |
| Inherent risk classification | Low risk or High risk, following your agency's assessment. A high-risk system needs whole-of-government approval as well as your agency's. Litmus warns you if you set a public-facing system to low risk. |
| Risks to test for | The risks you're concerned about, such as giving wrong eligibility advice or revealing personal data. Add your own, or choose Generate with AI and keep the suggestions that fit. |
| Knowledge base (optional) | The source documents your system answers from, such as the policy it explains: PDF, Word (DOCX), text or Markdown files. Litmus writes test cases grounded in them, and its judges use them to check whether answers are supported. It keeps the extracted text, not the file. A scanned PDF without a text layer gives no text. |
Next: Connect your system becomes available once you've filled in the name and description and confirmed at least one risk.
Connect your system
On Connect your system, set how Litmus calls your system's API. Connecting your system explains each setting. Then choose Run connection check. Finish onboarding becomes available once a check succeeds.
If an engineer will connect the system, add them as a project user first. Then choose Send to an engineer to copy a link to this page for them.
You can stop and come back later. A system you haven't finished adding stays in your list, and opens on its first page.
Add your team
Add teammates under Project users: choose Add user, enter the email address they sign in with, in lower case, and choose Editor or Viewer.
| Role | Who has it | What they can do |
|---|---|---|
| Owner | The person who added the system | Everything: edit the system, run tests, manage project users, decide what to do about issues, and submit the assurance record. |
| Editor | Teammates you add | Edit the system's details, criteria, personas and datasets, and run tests. They can't manage project users, decide on issues or submit the record. |
| Viewer | Teammates you add | View the system and its results. |
Litmus doesn't send invitations or email, so tell your teammates when you've added them. They sign in with TechPass, or Google if they're at MOE or GovTech, using the address you entered.
What good looks like
- The description says what the system is for, who uses it and what it must not do, in plain words.
- Each risk is specific enough to test. "Gives wrong eligibility advice" is easier to test than "inaccuracy".
- The risk classification matches your agency's assessment.
- The connection check succeeds, and the answer it shows is your system's reply, not "nothing matched".
If you're stuck
- Generate with AI doesn't respond: add the risks yourself, then carry on.
- The connection check fails: see troubleshooting.
- A teammate sees "You don't have access to this system": check that you added them as a project user, with the email address they sign in with, in lower case.
- You want to delete a system: you can't yet delete a whole system yourself. Ask us.
Still stuck? Ask us using the onboarding form (opens in a new tab), and say which step you're on.