Skip to main content

Get started with Litmus or Sentinel

Find out which product fits your AI system, what to have ready, and how to start.

Which product do you need?​

Pick the task that matches where your AI system is. You can use either product on its own, or both.

Common situations​

Your system isn't live yet
Start with Litmus. Test it against your own criteria and look into what fails before launch. If it will need guardrails once it is live, ask for Sentinel now too, so your engineers can add the checks before launch.
Your system is live
Start with Sentinel. Check users' prompts and your model's responses while the system runs, and decide in your application what to do with each result.
You want both
Sign in to Litmus, and ask for Sentinel through the onboarding form. Test with Litmus, and add Sentinel checks before launch so they're running when you go live.
Your system has changed
A new model, prompt or data source? Run your Litmus tests again and compare the scores with earlier runs.
You're not sure your system fits
Check what Litmus needs and what Sentinel needs below, or ask us.

How to start​

  1. Step 1: Sign in, or ask for Sentinel access

    Sign in to Litmus with TechPass and add your system. For Sentinel, fill in the onboarding form: choose Onboarding, tick Sentinel, and give your agency, your government email address and any teammates to add. Vendors: ask the government officer you work with to fill it in. Our team then registers your team and gives you an API key. People can sign in to the Sentinel dashboard with TechPass once they've been added to your team.

    To be confirmed: what teams new to Sentinel should enter for the form's required "Tenant Name" field.

  2. Step 2: Connect your system

    For Litmus, add your system's HTTPS API endpoint. For Sentinel, add calls to the Sentinel API in your application.

  3. Step 3: Run an evaluation or start checking text

    Run your first Litmus evaluation, or start checking prompts and responses with the Sentinel guardrails you choose.

To be confirmed: that you get an email confirming we have your form, and can reply to it.

Before you begin​

Who the products are for, what they cost, and the access and help you'll need.

Who they're for

Teams in the Singapore public service that are building or running an AI system.

To be confirmed: which agencies can use Litmus and Sentinel after the 19 October 2026 launch.

Cost

There's currently no charge for Litmus or Sentinel. Plans based on how much each team uses them will follow.

Litmus sends each test case to your system, so your system's usual running costs, such as its AI model usage, apply during tests. More about cost.

Approval

For Litmus, sign in with TechPass. For Sentinel, start with the onboarding form; our team then contacts you and helps set up your team's access.

To be confirmed: whether your team needs any approval, such as from your agency, before it starts.

Access you need

A TechPass account for each person who will sign in. For Sentinel, our team registers your team and issues your API key. You can't create Sentinel keys yourself.

Systems that work

Litmus: text-based AI systems, such as chatbots and assistants that answer questions from documents, tested one question and answer at a time. Your system needs an HTTPS API endpoint that can be reached from the internet. Litmus doesn't yet test multi-turn conversations, or image or audio input.

Sentinel: applications that send text to the Sentinel API over HTTPS. Sentinel checks text only, not images, audio or files. If your application runs on the Government Commercial Cloud (GCC) on AWS without internet access, ask us about connecting through a private endpoint.

To be confirmed: whether Sentinel accepts requests only from Singapore IP addresses.

Developer help

Yes, for setup. Someone who knows your system's API needs to connect it to Litmus, and an engineer needs to add Sentinel checks to your application.

Data you must not submit

Litmus and Sentinel are not yet cleared for CCE systems or data. Check with us before onboarding such a system.

To be confirmed: what CCE covers, which other data classifications each product supports, and what you must not submit.

Technical documentation

Start with the Litmus getting-started guide or the Sentinel getting-started guide. Inside Litmus there are also guided tours of each step.

What to have ready for Litmus​

Litmus tests your system by calling its API endpoint. Litmus checks the connection before you finish adding your system.

  • A TechPass account for each person who will sign in.
  • A short description of your AI system, the risks you're concerned about, which agency owns it, and whether it is low or high risk. Litmus can help you draft the description and suggest risks.
  • An HTTPS API endpoint for your system that Litmus can reach at a public internet address. Litmus can't currently connect to systems that are only reachable on an intranet or private network.
  • Any key or token needed to call your system. Litmus stores it encrypted.
  • The engineer who will connect your system, added to the system in Litmus first. Litmus doesn't send email invitations, so tell them when to sign in.
  • Optional: test cases you already have, as a CSV file, and reference documents your system answers from (PDF, Word .docx, text or Markdown).
Technical details for whoever connects your system
  • Authentication: a bearer token, an API key in a header you name, HMAC-signed requests, or none.
  • Timing: your system can reply straight away, or reply later by calling Litmus back.

The Litmus getting-started guide covers what Litmus needs from your system in detail. Inside Litmus there's also a short primer, guided tours of each step and a fictional example system to practise on.

Start with Litmus⁠ (opens in a new tab)

What to have ready for Sentinel​

Your application sends text to the Sentinel API and acts on the results.

  • A service on your server that can call the Sentinel API with the API keys our team gives you (one for staging, one for production). Keep the keys secret.
  • A policy: the checks to run, and the scores at which each one warns or fails. Your engineer creates it with the Sentinel API.
  • A decision on what to check: users' prompts before they reach your model, your model's responses before they reach users, or both. Sentinel checks whatever text you send.
  • Context for some checks: the off-topic and system prompt leakage checks need your system prompt (the instructions you give your model), and the refusal check needs the user's original prompt.
  • Logic in your application to act on the results. Sentinel tells your application what it found; your application decides whether to let the text through, block it or replace it with a safe message.
  • A TechPass account for anyone who will use the Sentinel dashboard. Your team's admin, or our team, adds you using the email address on your TechPass account.
Technical details for your engineer
  • Send your API key in the x-api-key header of each request.
  • Create a policy, the guardrails to run and the scores at which each one warns or fails, at https://sentinel.stg.aiguardian.gov.sg/api/v1/core/policies, and deploy it. Create it in production too.
  • Check text against the policy at https://sentinel.stg.aiguardian.gov.sg/api/v1/check in staging, while you build and test.
  • In production, with your production key, use https://sentinel.aiguardian.gov.sg/api/v1/check.

The Sentinel getting-started guide has code samples for your first request. The API guide and the guardrail reference have the detail.

Start with Sentinel⁠ (opens in a new tab)

What data will be processed​

Both products run on Amazon Web Services (AWS) in the Singapore region, in an environment GovTech manages. Some features send text to AI model services, which may process it outside Singapore. These include commercial models from providers such as Google and Microsoft.

Litmus keeps your system description, test cases, the text of documents you upload, and your system's full responses and their grades. To generate tests and grade answers, it sends your description, test cases, document text and your system's answers to AI models. During runs, it may also check test inputs and answers with Sentinel's guardrails, and Sentinel keeps a record of those checks too.

Sentinel keeps a record of each check, in staging and in production: the text you send, any context you include and the results, including any personal data a guardrail finds.

To be confirmed: how long Litmus and Sentinel keep your data, and how to ask for it to be deleted.

The FAQ has the detail: what Litmus keeps, and where it sends data, and what Sentinel keeps, and where it sends data.

Need help?​

Read the frequently asked questions, or ask our team.