Launched this week
Seldon
Behavioral QA agents that use your product like customers do
407 followers
Behavioral QA agents that use your product like customers do
407 followers
Find the tests you didn’t know to write. Tell Wren what should work in plain English. Its QA agents try the paths your customers take, like going back to change an address halfway through checkout or retrying a failed payment. When something breaks, you get a replay and the steps to reproduce the bug. First 10 runs free.
Free Options
Launch Team
Render Sandboxes: a computer for every agentIsolated compute, persistent state and a real filesystem
Promoted
Hey Product Hunt! I’m Tom, co-founder of Seldon. We built Wren because your entire test suite can pass and a customer can still find a bug five minutes after you ship.
We’re an applied research lab in SF. Our team includes researchers from MIT and Berkeley and people who’ve built products at Meta and Spotify. We put people and agents in front of the same screens and study where their decisions differ. That got us interested in a QA problem. A test can tell you whether the route you scripted works. Customers get to pick their own route.
How does Wren work?
Give it a goal in plain English, like “a new teammate should be able to accept an invite and join the workspace.”
Wren’s QA agents open your app and try different ways to reach that goal. One might open the invite while signed into the wrong account, switch accounts halfway through, and try the link again. You can test those paths without writing a separate script for each one.
When something breaks, you get a recording and the steps to reproduce it. You can run Wren alongside your existing tests in CI.
Who’s it for?
We built Wren for engineering and QA teams, from startups shipping new features every week to enterprise teams with years of test coverage. One early team told us Wren saved them days of QA before a release.
Try it on something you already test
Your first 10 runs are free. I’d start with a flow your team already has a passing test for. Did Wren catch something your suite missed? I’d love to hear what happened. We’ll be here all day.
Love this, @tom_charman!
Fun fact about how Wren started…
@marquise_rosier was on a customer call helping with something else when he spotted a problem nobody asked us to solve. He thought: we could probably build this. So we did.
A few weeks later, @dominic_kennedy shared some early feedback. Tests went from 10–15 carefully written steps to “go here, go there, click here. That’s it.” One design partner called it “a bit of a must.”
I remember that “wait, this could be something” moment.
Super excited we're opening it up for anyone to try. Can’t wait to see what folks break with it. 💛
@tom_charman The next big startup in San Francisco!
@dhaval_bhatt3 doing our best!
@dhaval_bhatt3 Thank you! its great to see all the support
@tom_charman I've been waiting for such a QA product for years. Back in the day, I hired non-engineer QA because I found they were better at breaking user flow and finding bugs that engineers wouldn't even consider.
NB: big fan of your branding, can you make a designer agent that creates experience users would like to have?
@samabrika1 Haha, thanks so much! That instinct for finding the one path nobody thought to test is exactly what we’re building into Wren. I’ll pass the branding love on to the team. And I should show you Finch at some point - it's built to make the lives of product teams and designers easier 👀
@tom_charman do I have a chance to verify the goal the agent understood?
@emil_bogomolov1 you do! you write down what should count as a pass before the run - and then you can watch the replay to check what the agent actually did!
@tom_charman nice approach to catching bugs that regular tests might miss
@spotch_founder Thanks Nick!
Congrats on the launch!!
our e2e hits a real BE, so we prefix everything it creates and sweep it after. an agent that picks its own path doesn't tell us up front what it made.
does a run clean up after itself, or do we point it at a throwaway account?
BTW, this is a great idea 👍
@seenew Appreciate the support - please do keep trying and keep telling us how it goes
@seenew this is an excellent question!! the run does not automatically clean up after itself, but you can add the cleanup as a step in your test case
@seenew or alternatively, as you mentioned, you can just point it to a throwaway acct -- depends on your preferences
@seenew Hello! The agent can be instructed to clean up after itself; you can just add this as one of the plain english steps! It's entirely up to you! :)
The "tests you didn't know to write" framing is what got me — as a solo dev I mostly hand-test the happy path and find the weird stuff from users after shipping. The half-finished onboarding flows are exactly where I'd point this first. One question: does Wren work on native apps too, or is it web-only right now? Most of what I ship is native, so that's the deciding factor for me.
@jcbuilds That's helpful to know! We're web only for now, but native is in the works and we’ll want early testers. If you're up for it, can we reach out when it’s ready?
Nice work Tom. We build crypto exchanges, and the hardest flows to test are behind KYC and 2FA, like signup to first deposit. Can Wren run on a staging setup with test accounts, and can I block any step that would move real money?
@skalmurad thanks! Yep, Wren can run on staging with test accounts. For signup to first deposit, you can tell the agent to stop before it submits the deposit. That instruction alone can’t guarantee a hard block, though. If you want to test the whole flow, I’d use a staging account that can only move test funds.
@tom_charman Thanks, that’s a fair answer. A staging account that can only move test funds is how we’d run it anyway, since an instruction is never a hard block. One follow-up: can Wren get past an OTP or 2FA step with a test authenticator seed, or do we need to turn 2FA off on staging for the run?
@skalmurad Yep, you can leave 2FA on if your staging account supports email codes. Each Wren agent has its own inbox, so it can pick up the code and enter it during signup. If your setup only accepts codes from an authenticator app, I’d need to confirm support for test seeds separately.
@tom_charman ok noted if Email codes work for us on staging, so that’s covered. Thanks Tom, good luck with the launch!
cool launch!! the hard part is knowing if a weird path it found is a real bug or just something no customer would ever do
@igorblink Thank you!! 100%! That’s why we built Wren on behaviour models of real users rather than random clicking, so the paths it tries are ones customers actually take. And every finding comes with a recording, so you can go in and watch back to verify. Would love to hear what you think if you give it a spin!
@tom_charman Very interesting, congrats on the launch! =)
Should I provide documentation about our product so your QA agents can test some real-life scenarios?
@adana thanks! No docs needed to get started. Just point Wren at your app and describe something a customer would actually do, like 'invite a teammate and check they can open the project.' If there’s a rule Wren should know about, pop that detail in the test too.
@tom_charman Thanks for the detils!
Wish you lots of success guys <3
@adana Thanks so much! Appreciate the support!
@tom_charman How does Wren deal with unexpected popups , loading delays, or interrupted user journeys ?