Kimi CLI from your phone, browser or desktop

Kimi is the one Happier will not upgrade for you, and that is worth knowing on day one rather than on the day a release goes out. Moonshot distributes it with an install script rather than a package, so there is no managed copy under ~/.happier/ for Happier to bump — the version on your PATH is the version you last installed, and getting a newer one means running Moonshot’s installer again yourself.

The second thing that is particular to this one is how a read-only session is made read-only. Kimi has no flag for it, so Happier writes a small Kimi agent definition of its own — one that extends Kimi’s default agent and removes its shell, write-file and string-replace tools — and starts the CLI with that file. The launch is kimi --work-dir pointed at the directory you chose, then acp, with --yolo added when you have turned approvals off. So Read-only here is enforced by the agent’s own tool list rather than by something outside it deciding what to veto.

Signing in is a slash command: the provider page starts kimi on the computer you picked and sends it /setup. Afterwards Happier does not go looking for confirmation — some of these status commands print tokens to a terminal, so the probe is left off and the page shows Unknown rather than reading something it has no business reading. If the session does hit an authentication error, that is where you find out: a 401 on the way out of the CLI is turned into a plain instruction to run kimi login and try again, instead of a stack trace in a log you were not watching.

Kimi CLI is Moonshot AI’s terminal coding agent, driving Kimi models against your repository. happier kimi runs the binary you already installed, in the directory you point it at, with your own Kimi account.

The part Happier adds is unglamorous and is the whole reason people install it: the session does not end when you leave the desk. Read it on a phone, answer a permission request, send the next instruction, and let it keep working on the computer that has the code.

The iOS and Android apps, the desktop app for macOS, Windows and Linux, and app.happier.dev are all clients onto the Kimi CLI process running on your computer — not read-only mirrors. Answer a permission request, send the next instruction, browse the repository, read the diff. The transcript is end-to-end encrypted before it leaves your computer, so the server carrying it holds ciphertext.

Happier installs on the computer that already holds your repository, and happier kimi starts Kimi CLI there as an ordinary subprocess, in the directory you point it at, signed in the way you signed it in. There is no gateway between the agent and Moonshot AI, and no copy of your source anywhere else.

Happier looks for kimi on your PATH and runs the copy you installed. Moonshot AI distributes it with an install script rather than a package; Happier can show you that command, and it will not execute a vendor install script on its own — vendor recipes are refused unless you explicitly allow them.

Moonshot AI’s own install and sign-in guide is at kimi.moonshot.cn/docs/cli. Happier does not mirror it, because a copy of someone else’s install instructions is a copy that goes stale.

Configuration reference for the Happier side — Kimi CLI in the Happier docs.

Kimi CLI is one of 13 command-line coding agents Happier runs, and they share one session list, one permission inbox, one set of keyboard shortcuts and one MCP configuration.

You define an MCP server once, and because Kimi CLI does not consume MCP directly, Happier hands the same tool surface to it through the happier tools shell bridge instead — the same tools, delivered as commands it can run. The app previews the effective tool surface before a session starts, so which one you got is visible rather than inferred.

That matters most on the days you are running more than one: sessions from different vendors side by side, and a notification tap opening the session that raised the request rather than the app’s front door. The full list is here.

Typing happier kimi in a terminal starts the session right where you are standing, and the same session is in the app at the same time. What you read in that shell is Happier’s own display of the session — the transcript, the permission prompts, the tool output — rather than Kimi CLI’s interface. Nobody has to give up the terminal to get a phone client.

There is no hand-off to Kimi CLI’s own TUI. That exists for Claude Code, Codex and OpenCode, and for Kimi CLI the session stays Happier’s on both ends. happier attach <session-id> lists the sessions on that computer and marks the ones it cannot reattach, rather than failing after you pick one.

The attach guide lists the cases instead of promising all of them.

One command on the computer that holds your code, then happier kimi in the repository you want to work in.

Walkthroughs are in the guides. All of it is MIT-licensed, and the relay is one you can run yourself.