Gemini CLI on iPhone: the 2 Ways That Work

You can run Gemini CLI on an iPhone, but not on the phone itself. Gemini CLI is Google's open-source Node.js agent, and iOS gives you nowhere to install it — there is no Termux for iPhone, and the on-device terminals that do exist cannot run a modern Node. So on an iPhone the agent always runs on a host elsewhere and your phone drives it. The fastest path is Cosyra on the App Store: a cloud Ubuntu 24.04 container with Gemini CLI already installed that you reach from a native iOS app. Paste a paid Gemini key, type gemini, and the agent is up. Sign up gets you 1 hour free, no credit card. The only other path that works is SSH from a terminal app into your own always-on machine, and the reason the obvious on-device options fail is worth understanding before you waste an afternoon on them.

This is the iPhone-specific companion to our Gemini CLI on your phone walkthrough, which covers iPhone and Android together. The iPhone earns its own page because iOS changes the answer in one big way: Android has Termux, so on a Pixel or Galaxy you can fight Node onto the device itself (we cover the Termux-plus-NDK path in Gemini CLI on Android). iOS has no equivalent, so the iPhone story is entirely about reaching a host. If you want the bigger tablet form factor with a Magic Keyboard, see Gemini CLI on iPad; for the full picture of every agent on mobile, start at our AI coding agents on mobile pillar, and Claude Code vs Gemini CLI lines the two agents up if you're still choosing.

Why won't Gemini CLI run locally on an iPhone?

Gemini CLI won't run locally on an iPhone because iOS has no Linux userland and the on-device terminals cannot run a modern Node.js. Gemini CLI is a Node binary distributed as @google/gemini-cli, and its README lists its targets as macOS, Linux, and Windows. iOS is not one of them, and Apple does not allow a Termux-style app that ships a general-purpose Linux environment with a package manager. So the two on-device iOS shells people reach for each hit a wall, and the failures are documented, not vague "compatibility issues":

• iSH crashes Node with "Illegal instruction." iSH runs Alpine Linux through a 32-bit i386 user-mode emulator on ARM hardware. Modern Node.js prebuilt binaries emit CPU instructions the emulator doesn't implement, so node --version dies with Illegal instruction — tracked in iSH issue #2335, open since January 2024. A second issue, #2604, is literally titled "Gemini cli" — a user asking how to install it, with no resolution. No working Node means no Gemini CLI.
• a-Shell ships no Node.js and forbids fork. a-Shell bundles native iOS-compiled and WebAssembly utilities, which makes it fast, but it doesn't include a Node.js runtime, and iOS sandboxing forbids fork(2) — the subprocess model the agent relies on to run shell commands. There's no apt or apk to add Node either. It's a genuinely useful Unix scratchpad on the go; it is not a host for an AI coding agent.

Here's the honest opinion the "just self-host everything" crowd pushes back on: if you bought an iPhone expecting it to double as a pocket Linux box for terminal work, the on-device story is worse than on a $200 Android phone, where Termux at least gives you a fighting chance. The fix isn't a cleverer install command: it's running the CLI somewhere that has a real Linux kernel. For the local-terminal limits in more depth see Termux for iPhone and Cosyra vs iSH.

What changed with Gemini CLI's free tier?

The price of entry to Gemini CLI changed on 2026-06-18. Until that day, you could sign in to Gemini CLI with a personal Google account and get a generous free quota (60 requests per minute, 1,000 per day) with no API key. Google retired that path — the no-key login now returns HTTP 410 Gone, and the official successor is Antigravity CLI. Per Google's own transition announcement, the open-source gemini binary keeps working, but only with a paid key from Google AI Studio or Vertex AI. Enterprise Code Assist licenses are unaffected.

So every iPhone path below assumes you'll paste a paid GEMINI_API_KEY. That's not a Cosyra opinion or a Cosyra limitation; it applies identically whether you run Gemini CLI in our container, on your own SSH box, or on a laptop. The binary itself is alive and shipping fast: the latest release was v0.49.0 on 2026-06-25, with about 106,000 GitHub stars as of 2026-06-30. Google took away the free first-party endpoint, not the toolchain. With Gemini 2.5 Pro selected on a paid key you still get the 1-million-token context window, which is the genuinely distinctive thing about this agent versus the others.

What are the two real ways to run Gemini CLI on iPhone?

There are two real ways to run Gemini CLI from an iPhone as of 2026-06-30: a cloud container with Gemini CLI pre-installed (Cosyra, native iOS app), or SSH from a terminal app into your own always-on machine. They're not interchangeable — one needs a machine of your own, one doesn't. We walk through both.

1. Cosyra (cloud terminal, Gemini CLI pre-installed)

This is what we build. The Cosyra iOS app gives you a persistent Ubuntu 24.04 x86_64 container with Gemini CLI, Claude Code, Codex CLI, and OpenCode already installed, 30 GB of storage, and hibernation that resumes exactly where you left off. We pre-install the four agents because setup friction is the thing that actually kills agent-driven mobile coding; nobody wants to debug a Node version or a crashing emulator on a phone keyboard. Because the container is real glibc x86_64 with Node 20+, the emulation and NDK problems that kill the on-device shells simply don't exist. There's no SSH tunnel to keep alive and no laptop that has to stay awake, and the same container is reachable from iPhone, Android, and web, so you can start on the train and pick it up in a waiting room.

• Works when: you want zero setup and Gemini CLI running on a real Linux host without owning one, plus the other three agents next to it.
• Breaks when: you have no internet — the container lives in the cloud, so there is no offline mode. That is a real trade-off against a local terminal.
• Cost: 1 hour free on signup, no credit card. Extend with a 10-hour, 7-day trial when you want more. After that, $29.99/month or $300/year. You bring your own paid Gemini key and Google bills you for tokens. See pricing.

2. SSH from Blink Shell into your own machine

The classic remote-into-your-own-box path. Blink Shell is a paid iOS terminal built around SSH and Mosh. Mosh matters on a phone: it keeps the session alive across the cellular handoffs and IP changes that drop plain SSH the moment you walk between cell towers. You SSH into a desktop, Mac, or VPS where you installed Gemini CLI with npm i -g @google/gemini-cli, paste your paid key on the host, start tmux so the session survives app switching, and drive gemini from there.

• Works when: you already keep an always-on Linux box or Mac and are comfortable with SSH plus tmux.
• Breaks when: the host sleeps, the home connection drops, or you do not want to keep a second computer awake just to give the phone a target. Chaining the phone to a machine in the closet defeats the portability that made you reach for the phone on the train in the first place.
• Cost: the Blink+ subscription plus whatever the host costs. Tailscale's free tier covers the personal networking if your box isn't already reachable.

How do you set up Gemini CLI on iPhone with Cosyra?

You set up Gemini CLI on iPhone in about three minutes: install Cosyra from the App Store, paste a paid Gemini API key inside the container, and type gemini. No npm install, no emulator, no SSH tunnel, no Mac. These are the exact steps we run on a fresh iPhone.

Step 1: Install Cosyra on iPhone

Open the App Store on your iPhone and search for Cosyra. Sign in with Apple, Google, or email and the app provisions a fresh Ubuntu 24.04 x86_64 container on first launch, about 15 seconds in our testing, with Gemini CLI already on the PATH. The four agent CLIs are inside, so you skip the part that breaks most on-phone setups.

Step 2: Add a paid Gemini API key

Since the free personal-account login retired on 2026-06-18, Gemini CLI now needs a paid key. Generate one at Google AI Studio (or use a Vertex AI key), then in the Cosyra terminal set it once so it persists:

The key lives with the container's home volume, so you paste it once and it's there next time you open the app. Google bills you for tokens directly — this is the same BYOK model we use for every agent in the container.

Step 3: Clone a repo and run gemini

Clone a repo and start a session:

Gemini reads the repo and starts working. When you close the app to catch your stop, the container hibernates; reopen it and the session is exactly where you left it, with no re-clone and no re-auth. On a phone that resume behavior is the whole game — you get pulled away constantly, and a setup that loses state every time you switch apps is unusable.

What does iPhone Gemini CLI look like in practice?

iPhone Gemini CLI in practice is short prompts, one hand, and dead time that used to be wasted. The phone is not where you do a four-hour refactor; it's where you keep a long task moving while you're away from the desk. Two patterns we run constantly:

The commute kickoff, dictated

Standing on the platform, you remember the failing CI job from last night. Open Cosyra, pull the branch, and hand Gemini the whole problem in one dictated sentence — tap the mic key on the iOS keyboard, say it, done. No hunting for symbols on a phone keyboard:

By the time the train arrives you have a diff to skim. The 1-million-token context on Gemini 2.5 Pro earns its keep here: you can point it at a large repo and ask it to trace a value across many files without first hunting down which files matter. Dictation plus a big context window is a surprisingly good fit for the phone, because the slow part on a phone is typing, and you've removed it.

The waiting-room review

Fifteen minutes in a waiting room is enough to review a teammate's pull request. Pull the branch, ask Gemini for a structured review of what you care about, read it on the phone, and post it with gh pr review --comment from the same terminal. The container holds your gh auth, so there's no login dance. This is the kind of work the phone is genuinely good at: bounded, reviewable, no deep window juggling. The single honest limit is the screen — a sprawling multi-file diff is more scrolling than on a laptop, so we keep phone tasks scoped to things you can actually read on a 6-inch screen.

How do the iPhone Gemini CLI options compare?

The options compare cleanly when you line them up against what actually matters: does Gemini CLI run at all, does it need a machine of your own, and what happens when that machine sleeps. The table covers the two working paths plus the two on-device shells people try first and why they fail, as of 2026-06-30. For the wider field of editors and terminals beyond Gemini, our coding on iPhone guide lines every option up, and best terminal apps for iPhone covers the SSH-client side.

Which path should you pick?

Choose Cosyra if you want Gemini CLI genuinely running with no machine of your own to maintain, the same container across iPhone, iPad, and web, and the other three agents next to it. Choose Blink plus SSH if your dev box is already always-on and Linux, and you want Gemini running on hardware you control. iSH and a-Shell are in the table only so you don't lose an evening discovering what their issue trackers already say. The one thing iOS cannot give you is Gemini CLI running on the phone itself; on Android the calculus shifts because Termux gets you closer to local, which is the whole point of our Gemini CLI on Android guide.

Frequently asked questions