OpenClaw, the self-hosted open-source AI agent, now has a native app for iPhone and iPad. Until now, controlling it on the go required workarounds built on messaging apps, but the dedicated app removes that hassle. You can now securely connect to the AI agent running on your computer at home or in the office straight from the iPhone in your hand.
Taking the AI Agent That Runs on Your Own PC With You
OpenClaw is a self-hosted AI agent that runs on your own Mac or Windows PC. Rather than handing everything over to a provider's servers the way a service like ChatGPT does, you connect your own API key from an AI service such as Claude, OpenAI, or Gemini to a "gateway" running on your local machine. From there, the agent can reach the files, messaging apps, and web browsers on that machine and carry out the tasks you give it. The appeal of the tool is that your data and processing stay within your own environment.
Previously, operating it from an iPhone or iPad while away from your desk meant relying on workarounds that routed through messaging apps like Telegram or WhatsApp. The new native app replaces that roundabout approach. Working alongside your existing gateway, it acts as a secure node that handles chat, approval of actions, sharing, and automation that taps into your device's features.
What You Can Do From an iPhone
Setup begins by pairing with the private gateway running on your home computer, using either a QR code or a setup code. Once connected, you can talk with the assistant from your iPhone and move along tasks waiting on the computer.
The app's main capabilities include the following.
- Chat with your assistant from your iPhone
- Use a voice mode called "Talk" that works in real time as well as in the background while the app is closed
- Review and approve actions the gateway is about to run, right from your iPhone
- Share text, links, photos, and other media from iOS directly into the agent
- Grant device features such as the camera, screen, location, photos, contacts, calendar, and reminders as needed
- Receive the status of connected workflows through push wakes and node status updates
Every one of these device features is designed so the agent can only use what you have explicitly permitted. Handing over something captured with the camera, or approving work on your computer from a distance, can all be done from a single smartphone. The app is available for free on the App Store, and a companion app for Android is offered as well.
The Trade-Offs Behind the Convenience
Powerful as it is, getting started and running it call for a fair amount of care. First, using the iOS app requires keeping a gateway running continuously on a machine at home or in the office, so it is not something just anyone can pick up casually. And because the agent operates on your files and apps, the device running the gateway has to be granted broad system permissions.
There is also a weakness common to AI agents in general: limited resistance to "prompt injection," in which malicious instructions are slipped in to hijack the AI. Precisely because this is a tool with deep access to your own computer and data, it is worth being deliberate about which apps and information you let it touch.
A Rename From "Clawdbot"
OpenClaw also carries a small backstory involving its name. It began life as "Clawdbot," created by developer Peter Steinberger. The name was a nod to the fact that the earliest version used Anthropic's Claude, but Anthropic took issue with it, which led to the rename to today's "OpenClaw." You could call it a very AI-era naming episode that surfaced as an open-source project grew popular.
Summary
OpenClaw's iOS app is a move to make a self-hosted AI agent usable even when you are not sitting in front of your computer. By connecting to the gateway with a QR code, you can now handle chat, voice, action approvals, and device integrations all from an iPhone. At the same time, the freedom of keeping everything in your own hands comes with responsibilities the user must shoulder, including running an always-on gateway, granting broad permissions, and guarding against prompt injection. For anyone who wants to entrust work to an AI while keeping their data within their own environment, it looks like an option worth trying.
