QU Studio turns the coding agents you already use into a team.Q, your project assistant, plans the work, hands it out, keeps watch, and only calls you to decide and sign off.
For indie developers, small teams, and anyone with something to build · Your own agents and account
One person. 10 days. 50+ releases.
One person, with Q leading Claude Code and Codex, from the first line of code to the launch of this site:
3,238 commits, 88% with an AI co-author credit
Figures from QU Studio’s own board and repo, as of Nov 2026.
Works with the agents you already use
One running example: in Café Orders, you say “add a category filter to the menu page.”
Q and the system handle the rest: breaking the work down, scheduling, watching progress, chasing, retrying, merging code, and remembering who knows what.
Q quotes your own words before acting for you, and everything it does can be undone.
Q never runs commands or edits files. It plans, delegates, and checks — the work happens in sessions you can open at any time.
Who it’s for
One person, run like a team.
Several Macs, several agents, one board.
Say what you want. Decide. Accept.
Prototypes
Versions
Each hand-in adds a line in English and Chinese, sorted into New, Improved and Fixed.
Every hand-in is test-merged with the branch it goes into; conflicts are fixed on the spot.
It merges into main first, then hands the release to the Mac with the signing certificate. Publishing is still up to you.
Sites and the like can go live with every change instead of waiting for a version.
Your agents
QU calls the agents you installed and signed in to, on your Mac. It never signs in for you or touches your login files, and usage counts on your own account.
Halfway through with one agent? Hand the session to another. What you discussed comes along.
Let one write the code and ask another to write tests or review it.
When one runs out, switch to another and the work doesn’t stop.
Helpers
Split the work, get a second opinion, keep going.
A session breaks a job apart, calls in a few helpers to work on it at once, and merges their changes back.
Code written by Claude Code? Ask a Codex helper to write the tests or review it. It works the other way too.
A review helper can only look. It finds problems and never touches your code.
Need a key or a tool that only another Mac has? Call the helper in over there.
Sessions working together
Look sessions up by project, task or keyword.
Sender, project and task are written by the system, so the other side knows how to answer and it can’t be faked.
A message slips into the other session’s current turn without stopping it.
Two projects line up an interface directly, and each changes its own side.
A session can wait on another task and get told the moment it finishes.
Messages between Macs go through a relay, end-to-end encrypted.
Multiple Macs
Load balancing for your agents. When a new task comes in, Q checks which Mac still has a free slot and which agent has the most quota left, then sends it there.
Even a rented machine can run your agents: your repo never has to live there. Access is granted per task, expires, and can be taken back any time.
Safe to hand off
While you’re away
Model presets
Set up a few presets once. Q picks one per task: model and thinking effort follow the work.
Which agent, which model, how much thinking effort, fast mode or not, plus one line on when to use it.
Copy and styling get the light preset, pages and features the everyday one, designs and hard bugs the strongest.
A session calling in a helper picks a preset the same way.
Name a preset and that’s the one used. Change them any time.
Quota
Presets match model and effort to the work (see Model presets above).
It stays below a quota line.
Move to another agent and bring the context along.
When the quota resets, the work carries on.
Under the hood
The background service is a single Rust process that manages every session itself, with no Node or SDK host process wrapped around each one. Many sessions at once stay light.
Close the window and the work keeps going. Close the lid and it sleeps; work picks up when it wakes. It doesn’t get slower or heavier over a full day.
It calls the Claude Code and Codex you installed and signed in to, and usage counts on your own subscription or account. It’s the same agent you run in the terminal, unchanged.
How QU thinks.
Yours. QU calls the agents you installed and signed in to, and never signs in for you.
Your projects and sessions stay on your Mac. Between your Macs, traffic is end-to-end encrypted; the relay can’t see the content.
Claude Code and Codex are built in, and Q hands work to them. 44 more agents in the ACP registry, such as Gemini CLI, Cursor and OpenCode, can run sessions too.
No. You can just say what you want, look at the prototype, and accept the result. Developers can open any session and take over.
An agent does the work in one session. Q plans it, hands it to several sessions, watches them and checks the results, so you only decide and sign off.
Windows and Linux versions are coming soon. Right now QU Studio is macOS only, on Apple silicon.
Closing the lid puts it to sleep, and work picks up when it wakes. To run overnight, send the work to a Mac that stays on, like a Mac mini.
Q picks the model and effort by the kind of work, so small jobs use lighter models. Background work keeps below a quota line so it doesn’t eat into yours, and when one agent runs low, work moves to another.
It doesn’t touch changes you haven’t committed; it stops and asks. Files outside git that can’t be regenerated are never deleted, and you see the list before any cleanup. Rollbacks can be undone, and commands that cross a red line are blocked.
macOS on Apple silicon only for now.