OpenClaw released native applications for iOS and Android on June 29, turning smartphones into secure remotes for the open-source AI agent platform’s self-hosted gateways. The apps connect to a user’s existing OpenClaw gateway over a paired connection, relaying commands, approvals, and notifications — all agent inference stays on the gateway hardware.
Pairing and Core Features
Setup requires scanning a QR code or entering a setup code to pair the phone to a gateway running on Mac, PC, or Linux hardware. Once paired, users can chat with their agent directly or switch to Talk mode for real-time voice conversations, according to Digital Trends’ Manisha Priyadarshini.
Every action the agent wants to take on the gateway still requires explicit user approval. Push notifications surface workflow status in the background, meaning users can approve or reject agent actions without keeping the app open.
The apps also support sharing text, links, and media into OpenClaw from other phone apps, plus selective access to device features: camera, screen, location, photos, contacts, calendar, and reminders.
Platform Differences
The iOS version requires iOS 18 or later, is free, and lists itself as a Productivity app that collects no user data per its App Store listing. The Android version requires Android 12 or higher. Early reviews note the Android interface is rougher, while the iOS app appears more polished.
Competitive Context
The launch follows a pattern NCT documented on July 2: persistent agent architectures where the runtime lives on servers and phones serve as lightweight approval interfaces. OpenClaw’s implementation is the open-source version of this pattern.
The timing is notable because Google is reportedly developing a competing “24/7 personal agent” designed to run continuously and handle tasks across devices. Google’s version would presumably run on Google infrastructure rather than user-owned hardware, creating a direct philosophical split: self-hosted open-source versus cloud-dependent closed-source.
The Self-Hosting Trade-Off
OpenClaw’s mobile strategy reinforces its core architectural bet. The phone stays a terminal throughout — agent inference runs on the gateway, not the handset. That means users retain full data sovereignty (nothing leaves their gateway unless they explicitly grant device access), but it also means they need to maintain gateway uptime themselves. If the Mac or Linux box running the gateway goes down, the mobile app becomes inert.
For users already running OpenClaw gateways, the mobile apps remove the last major friction point: needing to be at the same machine to interact with or approve agent actions.