AI agents move fast. Your product should keep up.
Lirris is a platform that continuously tests how your SaaS handles the AI agents your customers use, across different models and scenarios, and helps you investigate problems and verify fixes.
Your customers can put AI agents to work in your product.
Customers can delegate tasks in your product to AI agents through MCP, APIs, browser use or computer use, letting the agent decide how to carry them out.
An agent can act within its permissions and still get the task wrong, and if your product accepts that mistaken action, it can lead to unintended changes to customer data.
Agents tend to make the same mistakes.
Lirris tests how your product handles each one from the first run, starting where a mistake would cost the most.
- Plus the ones specific to your product.
Test how your product handles agent mistakes.
When Lirris finds a problem, you get the evidence to understand what happened.
Scenarios based on your product
Lirris generates relevant scenarios and varies things like the agents, tasks and starting data to explore how your product responds.
Findings you can work with
See why a failure happened and what it caused, with a suggested safeguard such as a confirmation step. The scenario is saved, so you can check your fix against it.
Lirris keeps checking as your product changes.
Your checks run automatically before every release. A small change can remove a safeguard by accident, like the confirmation step on refunds.
When the agents change
The agents your customers use get new models every few months, and a new model can handle the same task differently. Lirris tests your product again each time.
After you fix something
The scenario that found the problem stays in your checks, so if a later release brings it back, you’ll see it right away.
Run Lirris in your own environment.
Lirris is designed so that tests run on an open-source runner inside your environment, on a laptop or in your release pipeline. The runner uses your credentials locally, and only the results come back to Lirris.
- Staging or sandbox
- Testing starts in an environment your team chooses, using the accounts and data you make available.
- Local credentials
- You never hand your credentials to Lirris, because the runner reads them wherever it runs.
- Access you control
- Your team chooses which features and interfaces are available to each run.
- Actions you can review
- Recorded steps and resulting changes let your team see what happened during a run.
Common questions.
For anything else, email us at hello@lirris.com.
Is Lirris for testing our own AI features?
No. Lirris tests how your product handles the AI agents your customers use, such as ChatGPT or Claude acting on their behalf. Your product does not need any AI features of its own.
How is this different from our end-to-end tests?
Your end-to-end tests follow the steps your team wrote. Agents work out their own steps, and those can vary with the model or from one run to the next. Lirris gives the same task to different agents and checks what changed in your product.
Which agents does Lirris test with?
Lirris is being built to test agents that work through MCP, APIs, browser use or computer use, on different models. In early access, we’ll start with the agents your customers use most.
What does early access involve?
We’ll start with a call about what you’d want to test first, then set Lirris up with you as each part becomes ready. Early access costs less than Lirris will after launch, and we’ll agree the price with you before anything starts.
Get early access to Lirris.
Lirris is still in development. Apply to join the first group of teams and we’ll set it up with you. You’ll hear from us within two working days.
“People are handing more and more work to their agents. That’s why I’m building Lirris, to help prepare the internet for that future.”