Xiaolongxia OpenClaw launches iOS and Android mobile app: Remotely control Agent via phone on the go.

Open-source AI agent framework OpenClaw announced yesterday (30th) the official launch of iOS and Android mobile versions, with a core design of "local-first." The phone pairs with a user-hosted OpenClaw Gateway via WebSocket, allowing agents to compute on machines under your control.
(Previous context: Google NotebookLM launches 60-second short video feature: Long research papers turned into IG summaries)
(Background supplement: Fable 5, Mythos 5 are coming back! Anthropic officially announces relisting tomorrow, US Commerce Department lifts export controls)

Table of Contents

Toggle

• What is Gateway: The routing layer between the phone and agents
• What tools and permissions can the phone give to agents
• Agents moving to the mobile end

Open-source AI agent framework OpenClaw announced on X yesterday (30th) the official launch of iOS and Android mobile versions, continuing the framework as an open-source project.

The operational logic of the OpenClaw mobile app is fundamentally different from typical cloud chatbots. The model of mainstream assistants like ChatGPT and Gemini is: every sentence you input is sent to the provider's servers, processed, and then returned: data passes through third parties throughout.

OpenClaw takes the "local-first" route. Simply put, data and computation stay on your own machine, without detouring to external clouds.

OpenClaw is now on iOS + Android 🦞

📱 Native mobile apps, finally
💬 Agents in your pocket
🔔 Channels, tasks, replies on the go

Run agents from wherever your thumbs are.

iOS: https://t.co/7LHHc9htgM
Android: https://t.co/X0Wuh2uA8w

— OpenClaw🦞 (@openclaw) June 29, 2026

What is Gateway: The routing layer between the phone and agents

The key component that enables this architecture is called OpenClaw Gateway. Gateway is the routing layer—simply put, it acts as a relay, connecting the requests sent by the phone to the AI agents running on your computer or home server, as well as the tools and skills those agents need to call.

Gateway is self-hosted by the user, meaning you control the entry point for computation, rather than relying on some company's infrastructure.

The pairing method is deliberately low-barrier: scan a QR code or enter a setup code, and the phone establishes a connection with the Gateway via WebSocket. WebSocket is a protocol that allows continuous two-way communication between two endpoints—simply put, no need to re-handshake for each interaction, messages can flow in real time both ways.

What tools and permissions can the phone give to agents

After pairing, users can actively grant the agent access to native phone functions, including: camera, screen, location, photo library, contacts, calendar, reminders. This makes the agent no longer just a text-based Q&A tool, but one that can perceive your location, read your schedule, and operate your camera (with your active authorization).

The official demo page lists a wide range of use cases: from assisting with programming and querying data to planning meals based on your daily calendar, all with real-world examples reported by community members. What the agent can do essentially depends on how many tools you give it access to and what skill modules are connected behind the Gateway.

Also, because each permission is actively and individually granted by the user, control is returned to the user at this level: you can allow calendar access but not camera, or vice versa. The line of what the agent can and cannot access is drawn by you.

However, it's worth noting that based on early user experiences: the Android version received quite a few negative reviews at launch, with some users reporting app crashes, pairing issues with Gateway, and a bare-bones interface; but other users said the experience was smooth after setup.

The iOS version, which bypassed Apple's standard security review for distribution, sparked discussions among security researchers about sensitive permission boundaries. However, these frictions reflect the early stage of an open-source project moving from desktop to mobile, and are not fundamental flaws in functional design.

Agents moving to the mobile end

OpenClaw launching a mobile version is a slice of a larger trend. In 2026, the development axis of AI agents is shifting from desktop and cloud to mobile: not just a chat interface, but embedding deeper into the operating system.

Some developers have already integrated agents into input methods and keyboards, making every text input scenario an entry point for agents.

How much an agent can do depends on how many tools you're willing to give it. And the mobile version adds the entire phone's sensors and database to that answer.