Run OpenClaw on iPhone
Deploy and operate an OpenClaw agent end-to-end from iOS — Gateway on whichever host you own (or rent), chat client on your phone.
TL;DR
OpenClaw on iPhone means two things in 2026: pairing your iPhone as a node to an existing Gateway running on a Mac, Linux, or Windows machine — or using Onepilot to provision that Gateway yourself on whichever host you have available. The host is your call: a Mac mini in the closet, a Linux box at home, a Windows tower at the office, a Raspberry Pi 5, a NAS running sshd, or a $5/month VPS at Hetzner / DigitalOcean / AWS Lightsail. As long as you can SSH into it from your phone, Onepilot can install the Gateway there. The wizard takes about two minutes: pick the host, pick an LLM provider from 23+ supported, paste the API key, pick a channel (Telegram, Discord, or Slack), and confirm. The OpenClaw Gateway runs on the host; the iPhone is the supervisor.
Why OpenClaw needs a host, not just an iPhone
OpenClaw is built around a Gateway-and-nodes architecture. The Gateway is the long-running process that holds the agent loop, memory, skills, and outbound channels; nodes (iPhone, Android, browser, Mac) are clients that pair with it. The official iOS documentation states this explicitly: the iPhone connects to a Gateway as a node and does not run OpenClaw by itself.
That design choice is correct — agents need a stable host so memory and cron-like loops survive a phone going to sleep. The practical question for an iPhone developer becomes: where does that host live? Anywhere you can SSH into. Onepilot doesn't constrain the answer: a Mac mini you already own, a Linux box at home, a Windows machine on your network, a Raspberry Pi, a NAS, or a remote VPS — all are valid Gateway hosts. The wizard provisions OpenClaw on whichever one you pick, and the iPhone pairs with it as a node. Use your own hardware if you have it idle; rent a $5/month VPS if you don't.
Two paths to OpenClaw on iPhone, compared
| iPhone-as-node only | Onepilot wizard | |
|---|---|---|
| Run OpenClaw end-to-end from iPhone | Manual: install Gateway by hand on a Mac / Linux / Windows host | Wizard: provisions Gateway on any host you SSH into |
| Supported host hardware | Mac, Linux, Windows (you handle the install) | Same — Mac mini, Linux box, Windows, Raspberry Pi, NAS, VPS |
| Setup time | 30–60 min (CLI install + ngrok/Tailscale + pairing) | ~2 min wizard |
| Provisioning method | Bonjour or manual host:port + TLS fingerprint | SSH + nvm-scoped install (no sudo) |
| Pairing | /pair → /pair approve in Telegram | Built into wizard, no chat ping-pong |
| Background limits | iOS suspends sockets; canvas / camera / talk blocked when backgrounded | Agent runs on host; iPhone reconnects on resume |
| LLM provider choice | Configured on the host machine | 23+ providers from a wizard step |
| Channel routing | Manual Telegram/Discord setup per agent | Telegram, Discord, Slack — wizard step |
| Credential storage | Wherever the Gateway host puts them | iOS Keychain (Secure Enclave) |
Deploying OpenClaw from iPhone in four steps
- Add the host. Use any machine you can reach over SSH. Free if you already own one — a Mac mini, a home Linux box, a Windows machine with OpenSSH server, a Raspberry Pi, a NAS running sshd, a Docker container on your laptop, or a work machine reached through Tailscale. Don't have one idle? A rented VPS does the job: Hetzner CX11 ($4.51/month), DigitalOcean Droplet ($6), AWS Lightsail ($5), or Oracle Cloud Free Tier (free, ARM Ampere). Paste the hostname, port, and SSH credentials in Onepilot. SSH keys live in the iOS Keychain.
- Tap Deploy Agent → OpenClaw. The wizard opens the SSH session, drops a user-scoped nvm prefix (no
sudorequired, so it works on locked-down VPS images), and installs the OpenClaw CLI. No package manager surprises — the install is reversible withrm -rf ~/.nvm ~/.openclaw. - Pick an LLM provider and paste the key. Choose from Anthropic Claude, OpenAI, Google Gemini, Mistral, Groq, DeepSeek, xAI Grok, Perplexity, Cohere, Together AI, Fireworks, OpenRouter, Ollama, and others — 23+ providers as of April 2026. The key is stored on-device in the Secure Enclave and pushed to the server over the encrypted SSH transport.
- Pick a channel and confirm. Telegram is the fastest setup (paste a bot token, the wizard does the rest); Discord and Slack are equivalent. The Gateway starts, the iPhone pairs as a node, and the wizard reports back when the agent is live. From here you talk to the agent in the chosen chat app, in the Onepilot in-app chat, or by SSHing into the server directly.
What you get on the iPhone after deploy
An in-app chat tab. Talk to the OpenClaw agent the same way you'd talk in Telegram, but inside Onepilot — useful when you want to attach files from the iOS file browser or pull diffs from the git tab without leaving the app.
A real PTY terminal. Onepilot already runs the OpenClaw Gateway as a persistent background process — no tmux or screen ceremony needed. Use the terminal tab to SSH into the same host whenever you want to tail logs, watch the Gateway in real time, or run a quick command alongside it. The terminal is a SwiftTerm-backed VT100 emulator, not a command-list shim.
A file browser with syntax highlighting. Read the agent's working directory in 20+ languages — Swift, Python, TypeScript, Go, Rust, Ruby, and more. Diff and commit through the git tab when the agent edits code.
Cron jobs. Schedule a recurring run of the agent — daily summaries, weekly cleanup, hourly health checks — without writing crontab syntax by hand.
Channel routing. Telegram, Discord, and Slack become a second supervisor surface. Send a question from your watch, the agent replies in the channel, you read it without unlocking the phone.
What this does not replace
iPhone-side capabilities. OpenClaw's iPhone-as-node mode exposes camera, canvas, screen capture, and voice — those run on-device and are subject to iOS background limits. Onepilot does not change those limits; it changes where the Gateway lives. If you need the iPhone to be a sensor for the agent, pair it as a node alongside the server-hosted Gateway.
Telegram-only workflows. If you already run a Mac at home with the OpenClaw Gateway and you're happy talking to it via the Telegram bot, you do not need Onepilot. Onepilot is the path for developers who want the Gateway on a remote VPS and the iPhone as the only client they touch.
FAQ
Can I run OpenClaw on iPhone without a Mac?
Yes. OpenClaw is a Gateway-and-nodes architecture: the Gateway is a long-running process that lives on Linux, macOS, or Windows, and the iPhone connects as a node. Onepilot provisions that Gateway on whichever host you can SSH into — a Mac mini at home, a Linux box, a Windows machine, a Raspberry Pi, a NAS, or a remote VPS (Hetzner, DigitalOcean, AWS Lightsail, Oracle Cloud Free Tier). The host can be hardware you already own; it doesn't have to be rented. The Gateway runs on that host; the iPhone runs the client.
How does Onepilot install OpenClaw on a server?
Onepilot opens an SSH session to the server you choose, drops a user-scoped nvm prefix (no sudo required), installs the OpenClaw CLI, writes a config file, and starts the Gateway as a background process. The whole flow is a wizard: pick the server, pick an LLM provider from 23+ supported, paste the API key, pick a messaging channel (Telegram, Discord, or Slack), and confirm. Setup takes about two minutes per server.
Is the OpenClaw iPhone app on the App Store?
The OpenClaw team has published two iPhone clients (ClawOn and MyClaw) but both are positioned as node-only — they assume the Gateway is already running elsewhere. Onepilot is a different approach: a single iOS app that both provisions the Gateway on a remote Linux box and drives it. If you don't already have a Mac running OpenClaw 24/7, Onepilot is the path that doesn't require a second machine.
Where does OpenClaw actually run when I deploy from iPhone?
OpenClaw runs on whichever host you pick during the Onepilot wizard — a Mac mini, Linux box, Windows machine, Raspberry Pi, NAS, or any rented VPS (Hetzner, DigitalOcean, AWS Lightsail, Oracle Cloud Free Tier). The Gateway process and the agent's memory live on that host. Your iPhone stays the control surface: it sends commands, receives output, and routes to Telegram/Discord/Slack. If your phone is offline, the agent keeps running.
Can I drive OpenClaw from Telegram or Discord?
Yes. During deployment, Onepilot wires the OpenClaw agent to a messaging channel of your choice — Telegram, Discord, or Slack. After that, anyone in the channel (or the bot's DMs) can talk to the agent, and the agent's responses come back to the same place. Telegram is the most common pick because the bot setup is fastest, but the channel choice doesn't affect the agent's capabilities.
Does the OpenClaw agent persist when my iPhone is locked?
Yes. The Gateway is a long-running server process. Locking your phone, switching apps, or losing signal does not stop the agent; it keeps running on the VPS. Onepilot reconnects when you reopen the app, and Telegram/Discord/Slack messages are queued by the channel, so nothing is lost. iOS background limits only affect the iPhone-as-node features (canvas, camera, voice) — the core agent loop is server-side.
How does Onepilot wire an LLM provider key to OpenClaw?
OpenClaw reads its model from the providerKeyEnv set on the Gateway process. During the wizard, Onepilot stores your chosen API key in the iOS Keychain, opens an SSH session to the host, writes the key into a per-user environment file scoped to the OpenClaw process, and starts the Gateway with that env. The key never lives in plaintext outside the iOS Secure Enclave and the host's user-only file. To swap providers later, re-run the wizard step — Onepilot rewrites the env without touching the agent's memory or skills.
Run OpenClaw from your iPhone
Drop your email and we'll send one note when Onepilot ships on the App Store.
Join the waitlistLooking for Hermes on iPhone or the full agent index?