Claude Code, Codex, Gemini CLI and six more agents in a single window — no terminal required. Every conversation is kept and can be resumed where you left off. You can also drive it from your phone while you are away from your desk.
A free tier is available. Card payment via Stripe, cancel any time.
Please note: the app interface is currently in Japanese only. See Requirements before you buy.

Cursor and VS Code each ship an entire browser engine inside the application. That is why the app alone approaches 1 GB, why several processes start in the background when you open it, and why a few hundred megabytes of memory stay occupied the whole time.
ULCode uses the rendering engine that already ships with macOS. It carries no browser of its own, so the download, the disk footprint and the memory all stay small. You can keep using your other applications normally while the AI works. Even on a Mac with 8 GB of memory, you do not have to close things to make room for an AI editor.
All figures were measured on the same Apple Silicon Mac. Disk usage was measured with du; memory with the standard macOS footprint command, with ULCode and Cursor running at the same time. VS Code was not running at the time of measurement, so its memory and process count were not measured.
Whatever you said through Claude Code, Codex or Gemini CLI lands in one history. Search the full text by prompt, folder or model, read back through a session and press “resume” to continue that same conversation. You stop losing time hunting for what you asked last week.
Claude Code, Codex, Gemini CLI, Ollama, GLM, OpenCode, Kimi, Grok Build and Cursor CLI — nine agents you can switch between. You can also switch models mid-conversation. If you hold more than one account with the same provider, register each login once and swap with a single click. Remaining quota for each subscription stays visible at the bottom of the window.
Scan a QR code and you can send instructions to the AI running on the Mac at home. It works on the same Wi-Fi and from outside your network. Check progress on the move, approve steps and queue up the next instruction, and let it run. On the Mac version an optional setting keeps work running even with the lid closed, so you can check the result from outside. The connection is authenticated, so nobody else can use it.
Your local skills (~/.claude/skills) are listed, categorised and drawn as a graph; select one and it opens in the editor, ready to change. You can also have the AI write a new skill and save it for you. No need to remember where the configuration files live.
Projects in different working folders get their own tabs and run at the same time. A tab you closed can be reopened later. The list shows which ones are running and for how long. You can type the next instruction while one is still working, and it is sent automatically as soon as that run finishes.
Type “resize every png in this folder” in Japanese in the built-in terminal and it is turned into a command and run. The conversion happens on your Mac, so this feature costs nothing extra. Destructive operations such as deletion always stop for confirmation first. This feature currently expects Japanese input, and the explanation it shows is in Japanese.
A central progress view shows preparation, thinking, tool work, answer-waiting, completion and produced files. Answer waits glow, and the view counts down to the next stall warning or automatic-recovery decision. You can tune the automatic compaction threshold, and a silent run can perform one safety-gated automatic recovery.
Working folders you open are kept automatically as Spaces. Tasks in the same folder are combined into one clear state: Running, Waiting for reply, Possibly stalled, Queued or Stopped. You can see which job needs you and which one may be stuck without opening every tab.
Terminal sessions can be named and keep running when you switch tabs, hide the bottom panel or detach the view. Reconnect to the same session to restore its recent output. Ending the process is a separate two-step action, so closing a view cannot kill it by accident.
Tabs that paused waiting for an answer or a confirmation can be selected and resumed in bulk. Even with many jobs running in parallel, you can move just the stopped ones forward together.
When the AI offers choices and pauses, it turns the differences into a diagram and reads them aloud. Even without the jargon, you can see what each option will do before you decide.
The values and policies you write in your "self-axis note" are referenced by the AI in every tab, on every task. Agents that run outside your machine, such as GLM and Kimi, do not receive it by default. The twin feature (off by default) records what you fixed, chose or rejected, distils it into a notes file, and consults it when acting on your behalf.
API keys and login details are stored encrypted on your Mac. Taking one out requires Touch ID. You can hand a key to the AI's work without pasting it into the chat.
Apps and pages you build with the AI line up on a portfolio shelf you can revisit. Export them as a ZIP, or publish to GitHub Pages (requires the gh CLI to be set up).
Record your actions or grab a cropped screenshot from inside the app. Everything stays on your Mac, so no images are sent to an outside service.
These are real ULCode screens (the work shown is sample material prepared for the demo). The blue frames and numbers were added afterwards to point out what is happening where. Click any image for full size. The interface shown is Japanese, which is the only interface language today.




~/.claude/skills directly, so the skills you already use are the ones listed. Edit the file, save it, then call it as /csv-report and the AI produces the document.
cat sales.csv and runs. The conversion happens on your Mac, so the feature costs nothing extra.→ The table scrolls sideways
| ULCode Pro | Cursor | VS Code + extensions | CLI alone | |
|---|---|---|---|---|
| Monthly price (excl. tax) | ¥1,980 | about ¥3,000 | Free | Free |
| Several AI agents in one window | ◎ 9 agents | △ | △ (separate UI per extension) | × (launch each one) |
| Multiple accounts for the same AI | ◎ | × | × | × (swap env vars by hand) |
| Control from a phone | ◎ | × | × | × |
| Full-text history search and resume | ◎ across all agents | △ (its own history only) | △ (split per extension) | × |
| Installed size (measured) | 63 MB | 858 MB | 1.5 GB | — |
| Memory while running (measured) | 47 MB | 300 MB | not measured | — |
| Processes launched (measured) | 1 | 8 | not measured | — |
| Inference cost | your own subscription | included in the fee | depends on the extension | your own subscription |
| Interface language | Japanese only | English | English + localisations | English |
Memory and process counts were measured with ULCode and Cursor running at the same time on one Mac, using the standard macOS footprint command. VS Code was not running during the measurement, so those two rows were not measured for it.
Without a subscription and a completed setup for at least one agent, ULCode will start but will not be able to run anything. ULCode is an interface for the AI subscriptions you already hold, not an AI service of its own.
Team customers also need their own AI agent subscriptions (Claude Code and so on) for each user or for the company. AI usage is not included in the seat price. Invoicing and the quotation are set up for Japanese companies; if you are ordering from outside Japan, please ask us first. Enquiries in English are welcome.
First 50 seats · cancel any time · Japanese interface only