OpenClaw Launches Mobile Node Apps – Turn Your Phone Into an Extension of Your Self-Hosted AI Agent

Imagine this: you’re inspecting equipment on a job site. You speak into your phone: “Take a photo of the nameplate on the machine in front of me and log the current location.” Within seconds, your AI assistant captures the image, extracts text, tags it with GPS coordinates, and saves a timestamped work record.

That’s not a concept video. It’s what the newly released OpenClaw iOS and Android companion apps let you do today.

On June 29, 2026, the OpenClaw project published native mobile applications for both major platforms – available for free on the Apple App Store and Google Play. But these are not your typical “AI chatbot apps.”

OpenClaw architecture overview: Gateway as central control plane connecting chat apps and phone nodes

Phone as a Node: A Different Way to Think About Mobile AI

If you’ve used official apps from ChatGPT or Claude, you’re used to a familiar pattern: open app → type question → get answer. OpenClaw turns that model on its head.

The OpenClaw mobile apps are nodes, not standalone AI assistants. They act as peripherals that connect to a Gateway running on your own computer or server. Here’s a simple way to think about it:

  • The Gateway is the AI’s “brain” – it handles reasoning, planning, memory, and decision‑making.
  • The phone node is the AI’s “senses and limbs” – it provides camera, microphone, location, screen, and other hardware capabilities.

This separation means the app on your phone doesn’t contain an AI model at all. Its job is to expose your phone’s hardware as tools that the AI agent can invoke remotely. When you speak to your phone, the audio travels to the Gateway, where the agent processes it, plans a response or action, and then uses the phone node to carry out physical tasks like taking a photo or reading a notification.

You might ask: why not run the AI directly on the phone? That’s a fair question, and the answer lies in practical constraints. Phones have limited compute power and battery life, local models are far less capable than large models running on servers, and cross‑app permission handling is notoriously tricky on mobile. By decoupling the brain from the sensors, OpenClaw lets you run a full‑featured agent on a powerful machine while using your phone as an extension that gives that agent real‑world awareness – all without sending your private data to a third‑party cloud.

The Gateway: How the “Brain” Actually Works

To understand what the mobile nodes bring to the table, you first need to know what the Gateway does.

OpenClaw is an open‑source personal AI assistant/agent project, originally created by Peter Steinberger with contributions from a growing community. The core is written in TypeScript and runs on Node.js (Node 24 recommended, or Node 22.19+). The Gateway can be deployed on macOS, Linux, or Windows via WSL2.

Three Core Capabilities of the Gateway

1. It plugs into the chat apps you already use

OpenClaw doesn’t ask you to adopt yet another messaging app. Instead, it connects to your existing favourites: WhatsApp, Telegram, Discord, Slack, Signal, and iMessage. You send messages from those apps, they route to the Gateway, the agent processes them, and replies come back through the same channel.

This means you interact with the AI using tools you already know – no new interface to learn.

2. It performs real actions

The agent isn’t just a chat bot. It can:

  • Browse the web to retrieve information
  • Run shell commands on the host machine
  • Read and write files on your file system
  • Invoke hardware functions on connected phone nodes

3. It remembers and extends

OpenClaw keeps persistent memory across sessions, so it remembers what you’ve asked before. Community‑contributed skills and plugins let you add new abilities without waiting for a core release.

You Choose the Model

OpenClaw doesn’t bundle any AI model. You bring your own API key from a provider you trust – whether that’s a hosted service, a subscription‑based offering, or a local open‑weight model running on your own hardware. This “bring‑your‑own‑model” approach puts you in full control of inference costs, privacy, and vendor lock‑in.

How the Gateway and Phone Connect

Phone nodes connect to the Gateway over WebSocket, using port 18789 by default. The connection process balances ease of use with strong security.

Local Discovery: Automatic

When the phone and Gateway are on the same local network (Wi‑Fi), the app discovers the Gateway automatically via mDNS/Bonjour. No manual IP entry required – ideal for home or office setups.

Remote Access: Tailscale Recommended

For connecting while you’re away from home, OpenClaw recommends using Tailscale to create a secure VPN, then accessing the Gateway through a wss:// encrypted WebSocket endpoint.

A word on security: public endpoints must use WSS (WebSocket over TLS). Plain ws:// is limited to LAN and .local hosts, which prevents accidental exposure to the open internet.

Pairing: Explicit Approval Required

Every node must be explicitly approved before it can talk to the Gateway. The flow is straightforward:

  1. Open the app on your phone – it either auto‑discovers the Gateway or lets you enter the host and port manually.
  2. The phone sends a pairing request with its device identity.
  3. On the Gateway host, you list pending requests and approve the one you recognise.
  4. The node status changes to “paired · connected” and it’s ready to use.
# List pending device requests
openclaw devices list

# Approve a specific request
openclaw devices approve <requestId>

# Confirm the node is connected
openclaw nodes status

This ensures that no unauthorised device can join your agent network without your knowledge.

iOS Node vs Android Node: Feature Comparison

Both apps follow the same node philosophy, but each platform brings its own flavour:

Capability iOS – “OpenClaw – AI that does things” Android – OpenClaw node
Role Companion node Companion node
Pairing QR code or setup code Setup code or manual host/port
Chat interface Basic chat from iPhone Streaming replies, image attachments, full session history
Voice Realtime and background Talk mode Talk Mode (ElevenLabs or system TTS)
Canvas Canvas surface Live Canvas surface
Device capabilities Camera, screen, location, photos, contacts, calendar, reminders Camera, photos, screen capture, location, notifications, contacts, calendar, SMS, motion sensors
Action approvals Reviewable from the iPhone Managed on the Gateway
Data collection declared None None
Requirement iOS 18.0+ and a running Gateway A running Gateway on macOS, Linux, or Windows (WSL2)

iOS Highlights

The iOS app integrates with the system share sheet – you can send text, links, and media from any app directly into OpenClaw. For example, while reading an article in Safari, you can share it to the agent and ask for a summary, translation, or to‑do item.

Android Highlights

Android places extra emphasis on streaming chat replies and a live Canvas surface. Canvas is one of OpenClaw’s standout features – the agent can render dynamic dashboards, data visualisations, or interactive tool interfaces right on your phone screen, not just plain text. This is particularly useful for monitoring servers, tracking project progress, or managing real‑time data.

Android also supports text‑to‑speech via ElevenLabs or the system TTS engine, making voice conversations feel more natural.

Real‑World Use Cases: What Can You Actually Do?

Let’s walk through a few concrete scenarios to show how phone nodes add practical value.

Use Case 1 – Field Data Collection

A building inspector on site gives a voice command: “Photograph the third pillar on the east side, log the GPS coordinates, and check if last week’s rectification notice has been actioned.”

The agent executes a sequence:

  1. camera.snap – takes the photo
  2. device.location – captures the current GPS position
  3. Searches its memory for the previous rectification notice
  4. Compiles everything into a timestamped inspection record

All of this happens while the inspector keeps walking the site – no need to stop and type.

Use Case 2 – Location‑Triggered Reminders

The agent can set up geofences. For instance: “Remind me to check the parcel box when I get home.” As you approach your home, the agent detects the location change and pushes a reminder to your phone.

Use Case 3 – Notification Management (Android)

With the notifications.* command family, the agent can read incoming notifications and draft replies. When an SMS arrives, the agent reads it, suggests a response based on context, and sends it after your confirmation.

You can also set custom rules for email, messaging apps, and other notifications – the agent acts as a smart filter that reduces interruptions.

Use Case 4 – Live Dashboards on Canvas

Ask the agent: “Show me a live dashboard of my server health.” It renders a real‑time Canvas view on your phone with CPU, memory, and uptime metrics. The same works for stock prices, project burndown charts, or any data you can query.

Use Case 5 – Hands‑Free Conversation

Talk Mode supports continuous voice interaction – think of it like a two‑way radio with your agent. It’s useful when you’re driving, cooking, or exercising and can’t easily tap the screen.

One Important Constraint

Camera and screen capture commands require the app to be in the foreground. If the app goes into the background, those calls return an error. This is a platform security limitation, not an OpenClaw design choice – and it’s actually a reasonable privacy safeguard.

Privacy and Security: Built‑In, Not Bolted‑On

When an AI agent can access your camera, location, and contacts, privacy isn’t optional – it’s foundational. OpenClaw addresses this through multiple layers.

Permission Gating

  • Every hardware permission (camera, location, contacts, etc.) is requested through the standard mobile permission model – you grant them one by one.
  • “Privacy‑heavy” commands like camera.snap, camera.clip, and screen.record are disabled by default. You must explicitly enable them in the Gateway configuration.
// ~/.openclaw/openclaw.json
{
  gateway: {
    nodes: {
      allowCommands: ["camera.snap", "screen.record"],
    },
  },
}
  • A deny list (gateway.nodes.denyCommands) always overrides the allow list – so even if you enabled a command, you can still block it globally.

Pairing Credentials

  • Pairing credentials are stored on the device, and each connection requires manual approval.
  • The pairing record is a durable “role contract” – token rotation cannot upgrade a node to a Gateway or to a different role. Even if credentials leak, the attacker can only act as a node, not take over the entire system.

Transport Security

  • Plain ws:// is allowed only on trusted LAN and .local domains.
  • Public or Tailscale endpoints require a real wss:// TLS certificate.

No Data Collection

According to the privacy labels on both App Store and Google Play, the apps collect no user data. This aligns with OpenClaw’s self‑hosted, local‑first philosophy – your conversations, keys, and memory stay on your own machines. No data flows to OpenClaw servers because there are none.

Getting Started: A Minimal Walkthrough

If you want to try it yourself, here’s the complete step‑by‑step.

Step 1 – Set Up the Gateway

On a supported host (macOS, Linux, or Windows with WSL2):

# Install OpenClaw globally
npm install -g openclaw@latest

# Run initial setup and install the daemon
openclaw onboard --install-daemon

During onboarding, you’ll be prompted to enter your preferred model provider’s API key (Anthropic, OpenAI, or a local endpoint).

Step 2 – Choose Your Model

OpenClaw works with multiple providers. You’ll need an API key from your chosen one. If you’re using a local model, make sure it’s running and you have the correct endpoint URL.

Step 3 – Download the Mobile App

  • iOS: Search “OpenClaw – AI that does things” on the App Store.
  • Android: Search “OpenClaw” on Google Play.

Both are free.

Step 4 – Pair Your Phone

  1. Ensure the phone and Gateway are on the same Wi‑Fi (or configure Tailscale for remote).
  2. Open the app – it should auto‑discover the Gateway, or let you enter the address manually.
  3. The app sends a pairing request.
  4. Approve it on the Gateway:
openclaw devices list          # see pending requests
openclaw devices approve <id>   # approve the one you trust
openclaw nodes status           # confirm it's connected

Step 5 – Enable Commands as Needed

Edit the Gateway configuration file to allow specific node commands, then restart the Gateway service.

Step 6 – Start Using It

Now you can chat with your agent through your preferred messaging app or via the mobile app’s built‑in interface. Try simple commands:

  • “Take a photo with my phone and send it to me”
  • “Where am I right now?”
  • “Show a task dashboard on my phone”

Strengths and Limitations – A Balanced View

Strengths

Local‑first data control
Your keys, config, and conversation history live on your own hardware. No third‑party cloud, no unexpected data sharing.

One Gateway serves everything
A single Gateway handles multiple chat channels (WhatsApp, Telegram, Discord, etc.) and multiple phone nodes simultaneously. You don’t need to configure per‑platform integrations separately.

Hardware extension for AI
Phones give the agent senses it otherwise lacks – camera, GPS, motion sensors, notifications. This transforms the agent from a conversation partner into a capable assistant that can act in the physical world.

Privacy by default
The declared “no data collection” policy, combined with self‑hosting, gives you a level of privacy that’s rare among consumer AI products.

Limitations

The apps are useless without a Gateway
The mobile apps have no AI capability on their own. If the Gateway is offline or unreachable, the apps can’t do anything. This is a trade‑off you accept for the architecture.

Setup requires some command‑line comfort
You’ll need to install Node.js, run npm commands, edit a JSON config, and understand basic networking (mDNS, WebSocket, Tailscale). The documentation is clear, but it’s not a no‑click experience.

Foreground restriction for camera/screen
Because of mobile OS security, camera and screen capture only work while the app is in the foreground. This limits certain automation scenarios, but it’s a necessary safeguard.

Early‑stage project
The Android version shows a small download count, and the ecosystem is still growing. You may encounter rough edges or missing features.

Broad system access requires careful configuration
The agent can run shell commands and access files – that’s powerful, but you need to thoughtfully configure allow/deny lists to avoid unintended actions.

Frequently Asked Questions

Is OpenClaw really free?
Yes – the Gateway and both mobile apps are open‑source and free. You only pay for the AI model API usage if you choose a paid provider.

Do I need to be a developer to use it?
Not a developer, but you should be comfortable with basic command‑line operations (running npm, editing a JSON file, using a terminal). The project provides detailed step‑by‑step guides.

Can I connect multiple phones to one Gateway?
Absolutely. One Gateway can serve many nodes simultaneously, each with its own pairing approval.

Does the iPad version exist?
The iOS app requires iOS 18.0+ and runs on iPad, though the interface is optimised for iPhone. Check compatibility notes before installing.

How much does the model API cost?
That depends on your provider and usage. Light users might spend a few cents per day with providers like Anthropic or OpenAI. You can also run a local open‑weight model at no API cost, just the hardware overhead.

Can I revoke a device later?
Yes. Use the Gateway CLI to list paired devices and revoke access at any time. The device will no longer be able to connect.

Do the apps collect my location data?
No – the apps themselves do not collect or transmit location data. Location is only accessed when the agent explicitly needs it (e.g., you ask “Where am I?”), and only after you grant the permission. All data stays on your Gateway and phone.

Final Thoughts

OpenClaw’s mobile node apps represent a deliberate alternative to the mainstream cloud‑centric AI model. Instead of building a “super app,” the project turns your phone into an extension of a self‑hosted agent – giving that agent physical capabilities while keeping your data under your control.

Whether this fits your workflow depends on your needs and your comfort with self‑hosted tools. If you value data sovereignty, don’t mind a bit of terminal work, and want an AI that can genuinely interact with the real world through your devices, OpenClaw is a compelling option.

The project is still early, but its privacy‑first design and active community make it one to watch for anyone who believes that AI should work for you – not for a cloud provider.

Resources: