Most AI assistants are limited by one boring thing: someone has to build the integration. Cyclon is trying to skip that step entirely, and it’s selling a phone to prove it works.
The Y Combinator F26 startup, founded by Adam Zinebi and Thomas Kiefer, launched an AI phone OS on November 3, 2026. The first units ship as a Google Pixel 8a with Cyclon preinstalled, delivered within 30 days of preorder.
How it controls your apps
Siri, Google Assistant and Alexa-style tools depend on APIs and developer-built hooks. If an app doesn’t support the assistant, you’re out of luck. Cyclon works at the interface level instead. It taps through screens, reads on-screen elements and completes multi-step tasks across apps, the same way a person would.
The example given is finding photos in a WhatsApp thread and forwarding them to a friend. That’s a simple request, but it crosses app boundaries in a way most assistants still handle badly.
When no app exists for the job, Cyclon generates a mini-app on the spot. These run in the background while the phone sits in your pocket. It’s a neat idea, and it sidesteps the short list of services most assistants actually support.
Why the Pixel 8a
Cyclon operates almost entirely at the OS level, so it doesn’t need special chips. Starting on existing Pixel hardware lets the team ship in weeks, not years, and collect real usage data right away. That’s a smart contrast with AI hardware startups like Humane and Rabbit, which built custom devices first and then struggled to justify them.
Custom Cyclon hardware is planned for later. The goal is to run larger models on-device and call cloud models only when needed, which should help with privacy and latency. No timeline has been given, so treat it as a roadmap item, not a product.
The hard part is reliability
Screen-driven agents have a known weakness. Apps have different layouts, update constantly and hide edge cases everywhere. A button moves, a popup appears, a login expires, and the workflow breaks. This is the same problem facing computer-use agents from Anthropic and OpenAI, and those teams have far more resources and still ship them with caveats.
Cyclon has an advantage in owning the whole OS, which should give it deeper access than a third-party agent running on top of Android. But a demo that works is not a product that works on the thousandth app. Everything depends on the long tail.
The launch post drew about 704,000 views, so interest is real. Interest is cheap, though. Preorders and retention will tell more.
Does it matter?
Yes, with a caveat. The direction is right. Agents that use software as it exists today will beat assistants that wait for every developer to cooperate. Google and Apple are both heading this way, and a startup that gets there first on real hardware forces the question of how much control users will hand over.
But a small team is taking on the hardest version of the problem. Early buyers will get a real test within a month. If Cyclon holds up across messy everyday apps, big platform vendors should pay attention. If it doesn’t, it joins the list of AI phones that looked better in the launch video.
