Tiller is sunset. Every customer server terminated, every subscription cancelled, the production database backed up, verified, and wiped. Closed out cleanly, the way an ending should be.
Running a venture studio means placing several bets at once. That only works if you can also fold. Knowing what to build is half the job; knowing when to say no, or no more, is the other half. This is the story of a no more: why we built Tiller, why we ended it, and what comes with us.
The problem we saw
Tiller was born in the OpenClaw craze. Suddenly everyone wanted an autonomous agent of their own, and the options for actually running one were rough: a bare VPS exposed to the open internet, a physical machine humming somewhere in your house, or, and people really did this, your own personal computer, handing an autonomous agent control over the same machine that holds your passwords, your files, your life.
Tiller was our answer: a managed, European-hosted environment for running AI agents safely. Isolated, secure, one click instead of a weekend of server hardening. Every server came with OpenClaw preinstalled and ready to run. We shipped it and put it in front of real users with real workloads.
The test we set ourselves
Tiller was never only a product. It was also a test of Ryde itself: could a small studio take an idea, build something real around it, and ship it fast? Not a prototype or a landing page, but a working platform with provisioning, billing, and paying customers, live within days.
We could, we did, and that answer stays answered whatever else happened next. And people trusted a bare machine to us and let us turn it into something of theirs, which is the part we're most proud of.
The market moved
Then two things became clear, and neither was going away.
First, we were supporting the wrong thing. We stood fully behind Tiller, and supporting our platform was never the issue. But what users actually needed help with was OpenClaw itself: configuring it, understanding it, working around its quirks. That was not in our control, and it was never what we set out to do. Tiller was deliberately simple and affordable; the hands-on guidance OpenClaw users really needed wouldn't have been an add-on to that product, it would have been a different product entirely.
Second, we were building on top of a community-built product whose roadmap wasn't ours. To OpenClaw's credit, the maintenance burden stayed light. But every step toward tighter platform integration would have tied Tiller's future more firmly to decisions made elsewhere.
Our bet was that the missing piece was infrastructure: give people a safe place to run an agent, and adoption would follow. What we learned is that the bottleneck wasn't the infrastructure around AI. It was AI adoption itself.
Saying no more
We've written before that discipline at Ryde means killing work early when the signals turn negative. That sentence is easy to publish and harder to live. A studio that can't close things ends up dragging every half-alive bet behind it, and pays for that in focus.
Once the picture was clear, we had two options. Let Tiller quietly degrade while our attention drifted elsewhere, the slow fade most products get. Or end it deliberately: announce it plainly, give every customer time to move, wind down support, and shut it down properly.
We chose the second. We announced the wind-down in June, said exactly why, and closed it out the way we'd want a vendor to treat us. No zombie product, no surprise invoice, no data left lying around.
What comes with us
The lesson turned out to be bigger than the product.
Running an agent is not the hard part of AI adoption. Compute is a solved problem; we solved our slice of it and it wasn't enough. The hard part is everything around the agent: what it knows, what it's allowed to do, and who answers for what it just did. And just as much, it's the human side: knowing what to try, and feeling secure enough to try it. Handing real work to an autonomous system is scary, and no amount of server isolation makes that feeling go away on its own.
What we keep is AI infrastructure knowledge, and that's not a consolation prize. How to isolate agent workloads. How to make deployment one click instead of forty. How to meter usage cleanly and bill it by the day. How to run European-hosted systems people can trust. That's the muscle Tiller built, and it's the muscle everything we build next starts from.
Fair winds
A tiller is the smallest part of a boat and the only one that decides where it goes. Ours pointed somewhere the wind didn't follow. But we learned to sail by holding it, and we learned it together. That knowledge doesn't get terminated with the servers.
Thank you to every person who trusted us with a bare machine. And thank you to everyone who built it.
Fair winds.
