liana-cli

Connect a machine to Liana so she can work on it — your laptop, a work server, anything you own.

When Liana needs that machine (your files, your apps, a repo checkout), she moves there, does the work natively — the same tools Claude Code has, running as you, starting in the directory you connected from — and comes back when she's done. You keep talking to her wherever you already do: iMessage, Telegram, Slack, or the web.

Use

Get a pairing code from useliana.com → Devices → Add a device, then, in the directory you want her to start in:

npx liana-cli connect --code YOUR_CODE

Leave it running — it holds a connection in the background and reconnects on its own. Next time, no code is needed:

npx liana-cli connect

Options:

• --name "MacBook" — what the device is called in your dashboard (default: hostname)
• --dir <path> — the directory she starts in (default: the current directory)
• --code <code> — pairing code, first time only

What it can do

Out of the box: your files and your shell, wherever you connected it. That needs no permissions and is all most people want.

Anything beyond that is opt-in, one capability at a time:

npx liana-cli enable messages    # let her read and send your Messages
npx liana-cli status             # what she can and can't do here

enable walks you through the macOS permission and tells you exactly what to click. Nothing is requested unless you ask for it by name — connecting a machine and letting Liana read your texts are two different decisions.

There's no separate on/off setting to keep in sync: the macOS permission is the switch. Revoke it in System Settings and the capability disappears, so she stops offering what she can't do.

Messages (macOS)

Reading your messages needs Full Disk Access — the one permission an app isn't allowed to ask for, so you grant it by hand, once. liana enable messages opens the right screen and names the program to add. macOS only applies a grant to programs started afterwards, so there's a restart at the end of it.

Which program it names depends on where the work happens. With a background agent installed (liana connect --install) that's Liana.app, and you finish with liana connect --restart; without one it's your terminal, and you quit it with Cmd-Q, reopen, and run the same command again.

That last run also sets up sending, which needs a separate permission to control Messages. Often it's already granted and nothing happens; if not, macOS asks and you click Allow.

Optionally allow Contacts too and she'll say "Ray" instead of a phone number.

Once it's on, you can ask her things like "what did Ben say about Friday?" or "text Ray that I'm running 10 late" from any channel — she reads your Mac without you being at it.

Messages she sends go from your own number, as you. She'll use your own history to work out who you mean, and asks only when it's genuinely unclear.

Group chats work too — by name ("text the Offsite crew that I'm running late") or by naming the people in it. She can't yet see photos or attachments; those read as empty.

Checking what she sees

npx liana-cli debug                  # recent conversations, as she sees them
npx liana-cli debug "Ben Peng"       # one conversation: which handles matched
npx liana-cli debug send "Ben Peng"  # where a send WOULD go — sends nothing

These print names, times and message lengths — never message text — so you can compare against Messages.app without anything sensitive ending up in a terminal log or a paste.

Notes

• The pairing code is exchanged once for a device token stored at ~/.liana/config.json (mode 600). Revoke a device any time from the dashboard; the connection drops immediately.
• She runs as you on this machine, with your access. That's the point — but it means only connect machines you're comfortable with her using.
• The model credential stays on Liana's backend: this agent talks to the model through an authenticated proxy, so no API key is stored on your machine.