Built for agent workflows

Four things a terminal never had to do until now

Terminals were designed for a person typing one command at a time. Agents work in parallel, run long, and stop to ask permission. Ultenix is built around that.

● waiting on you

● working　42 / 118

✓ done

3 agents, one window

Per pane agent state

Every pane knows whether its agent is working, finished, or blocked on your answer. Tabs carry the same marker.

▲ Notification

claude needs your approval

"Overwrite these 3 files?"

tab: marketing site

Notifications that stay quiet

Nothing fires while you are looking at that pane. You only get pulled back when the agent is genuinely blocked.

Resume yesterday?

claude ・ ~/site

codex ・ ~/api

✓ reopened in place

Sessions that survive a quit Pro

Ultenix remembers what each pane was running and brings it back, continuing the Claude Code conversation rather than starting over.

⚠ execution paused

312 files would be deleted

rm -rf ~/Projects/old-site

[ cancel ] [ inspect ] [ run ]

A stop between agent and disk

Destructive commands are caught before they run, described in plain language, and held until you approve them.

What we redesigned

The three problems that show up every single day

None of these are the agent's fault. They are what happens when agent output lands in a terminal designed for a human typist.

Working with agents

A conventional terminal

Ultenix

An agent is blocked on approval

You find out when you look

The tab is marked and you get notified, on your phone if you are out

You quit, the context is gone

Brief the agent from scratch

Reopens where the session left off

You step away from the Mac

Everything waits until you are back

Answer from iPhone or iPad and it keeps going

iPhone & iPad

Approve from wherever you are Pro

Agent work is mostly waiting, punctuated by a question you have to answer. Answering it should not require being at your desk.

claude ・ marketing sitewaiting

codex ・ apiworking

zsh ・ photoson the Mac

merge them, keep the larger amount

Pair by scanning a QR code

Scan the code on your Mac and you are connected. On the same network it links directly; away from home it goes through a relay.

End to end encrypted

Keys are derived from your pairing phrase, so the relay carries ciphertext it cannot read.

Read the screen and reply

See every live terminal, scroll the output, send input, and start a new terminal on the Mac from your phone.

A day with agents

Hand off in the morning. Answer from a train.

09:20

Mac

Four handoffs

One job per pane, each with its own agent and folder.

claude ・ codex ・ claude

3 running

10:05

Lid closed

No babysitting

Ultenix stays quiet while you can see the pane, and marks the tab when you cannot.

tab markers

amber = blocked, green = done

11:04

Out ・ iPhone Pro

One question, answered

Only approvals reach you. Reply in a line and the agent continues.

Ultenix ・ approval3 entries look duplicated. Merge them?

14:30

Mac Pro

Back in place

Reopen and both sessions come back where they were.

✓ resumed

claude ・ codex

Fits your existing setup

A real terminal underneath

Nothing here is a wrapper around a chat box. It is a full terminal you can move into.

Panes and tabs

Split any way you like, drag panes between tabs, and save named layouts with their working directories.

Import from iTerm2

Bring your saved window arrangements across, directory by directory.

Make it yours

11 color schemes, any installed monospace font, per folder backgrounds, and 23 remappable shortcuts.

Find and reuse

Search the scrollback, copy the output of a single command, and open URLs and paths with a click.

No terminal expertise required

Or skip the command line entirely

Type what you want in plain language, or say it out loud. Ultenix writes the command, explains it before it runs, and stops at anything destructive. Useful when you are onboarding someone who does not live in a shell.

"resize every photo to 1200px"

sips -Z 1200 ~/Pictures/*.jpg

Resizes 38 files in place.

Plain language input

Always previewed in plain language before anything runs.

● recording…

"sort these invoices by month"

✓ transcribed on device

Dictation Pro

Speech is transcribed on your Mac. Audio never leaves the machine.

Setup 3 / 5

Installing Claude Code

npm i -g @anthropic-ai/claude-code

✓ installed

Guided setup Pro

Walks a newcomer from a clean Mac to a running agent, one step at a time.

Recipes

・ resize images

・ merge PDFs

・ set up Claude Code

A palette of common jobs

Pick a task, fill in the blank, and see the command before it runs.

Safety

Agents get to propose. You get to approve.

This app runs commands, touches your files, and talks to your phone. Here is exactly what that means.

⚠ Execution paused

This would delete 312 files.
It cannot be undone.

rm -rf ~/Projects/old-site

CancelInspectRun it

Signed and notarized

Distributed with an Apple Developer ID signature and notarization by LIFE SCAPE, K.K.

Dictation stays on device

Transcription runs locally. No audio and no transcript is uploaded anywhere.

Remote access is end to end encrypted

Your Mac and your phone hold the keys. The relay only forwards ciphertext.

Your own model, if you prefer

Point Ultenix at your Anthropic key or a local Ollama model and nothing routes through us.

Pricing

The terminal is free forever

What we charge for is the hosted AI and the link to your other devices.

Free

Free forever

Tabs, split panes, saved layouts, import from iTerm2
Search, colors, fonts, backgrounds, key bindings
Notifications for waiting and finished work, confirmation before risky commands
Bring your own Claude key, or run Ollama on your Mac

Download

Pro

$7/ month

$70 per year (two months free)

AI with nothing to set up. No API key, no model download
Speak your instructions
A guide that walks you through setup one step at a time
Resume the work you left
Drive your Mac from iPhone and iPad

Try 14 days free

See the full pricing page →

FAQ

Questions worth answering

Is the free version actually usable?

Yes. Tabs, splits, layouts, search, colors and notifications never expire. Connect your own Claude key or a local Ollama and the AI input works on the free plan too.

What is the "AI with nothing to set up"?

We host the model for you. There is no API key to obtain and no model to install.

Do I need my own Claude Code subscription?

Yes. Ultenix is where Claude Code runs; the Claude Code subscription itself is between you and Anthropic. Ultenix walks you through installing it.

Is my voice sent anywhere?

No. Transcription runs on your Mac. The first run downloads a small model file (about 57MB).

Can I move over from iTerm2?

You can import saved arrangements (how windows are split, and their working directories) and then save and restore them as Ultenix layouts.

What do I need to run it?

macOS 11 or later, on both Apple silicon and Intel. The iPhone and iPad apps are not published yet.

Rebuilt from the ground up for AI agents.