Changelog
What changed in each version of Cuewire, newest first. On Windows, Cuewire offers each new version in the sidebar and installs it when you restart.
0.1.23
- A new look. Cuewire is set in its own typefaces, keeps one violet accent for what needs you, and draws its icons instead of using text symbols. The sidebar, the canvas and every settings sheet were redone with calmer controls, in dark and light.
- Everything is reachable from the keyboard. Tab moves between the controls of a sheet and stays inside it; F6 moves between the sidebar, the canvas and the canvas tools. A ring shows where the keyboard is.
- A small node still tells you an agent is waiting: its badge shrinks to a dot instead of disappearing.
- Saving a note again starts on the file it was saved to last, instead of the project folder.
0.1.22
- Test the merge, in a layer's menu: Cuewire builds and tests the layer merged with your project and the other layers, without touching any of them. When it fails, a note says which layer, or which two layers together, broke it.
- Review changes also shows what the other layers changed, and the reviewer looks for changes that only work together, even in different files.
0.1.21
- Keep a note's text in a file of its own: right-click the note and choose Save as file…, or select it and choose Save note as file… in the command palette. It is saved as Markdown, named after the note.
0.1.20
- Remove a canvas or a whole project with Remove canvas… or Remove project from Cuewire… in its right-click menu in the sidebar. Your project folder stays as it is.
- Agents wired to a Shell terminal know how to run a command in it and read what it printed.
- Agents pipe command output into a note, so quotes and other characters arrive unchanged.
0.1.19
- A remote project keeps its canvas's notes on the remote computer, so agents there read and write the same notes you see.
- Agents on a remote computer can run
cuewireand get the Cuewire skill, as they do on yours. - On a remote computer, Cuewire knows when a Codex agent ends its turn, and Copilot CLI gets the canvas brief and its role.
0.1.18
- Agents on a remote canvas no longer stay on Starting…: Cuewire finds them by name on the remote computer.
- When a remote computer runs an older Cuewire server, Cuewire updates it. If terminals are running there, the canvas offers to restart them.
- A terminal that could not start says why, and Restart tries again.
0.1.17
- Agents installed in
~/.local/binon a remote computer now show up when you add an agent.
0.1.16
- Agents you install on a remote computer after connecting show up when you add an agent. You don't need to connect again.
0.1.15
- Type an SSH password once per remote computer, not once for each step of connecting.
- On Windows, password fields turn off input methods such as Chinese and Japanese ones, which changed what you typed. The SSH sign-in warns when a password has characters a plain keyboard doesn't type.
- When a remote computer's host key has changed, Cuewire says in plain words what that can mean and how to check it.
0.1.14
- Open a remote project on another SSH port: add it to the address, as in
user@host:2222:/path.
0.1.13
- A Settings sheet (Ctrl+,, or Cmd+, on a Mac) with links to every other kind of setting.
- Move Cuewire's data folder to another drive from Settings. Cuewire restarts, copies and checks your data, and opens again.
- New layers can go next to the project, on the same drive.
- Downloaded updates are deleted once they are installed.
0.1.12
- The list of agents under New agent here stays open while you move the pointer to it.
0.1.11
- The mouse wheel scrolls agents with a full-screen view, such as Claude Code's, even on a keyboard without Page Up and Page Down.
- On a Mac, Cmd+Shift+Y and Cmd+Shift+N answer a waiting agent even while a terminal has the keyboard, and shortcuts are shown with Cmd.
0.1.10
- Share files with a canvas's agents: drop them on the canvas, or paste a screenshot or copied files into a terminal. The agent gets the files' paths.
- Shared files show on the canvas as cards, and pictures as pictures. Double-click a card to open the file.
0.1.9
- Codex agents in Codex's elevated Windows sandbox can run
cuewire, so they reach the agents and notes wired to them.
0.1.8
- Notes keep their newest lines in view, so lines an agent adds show as they arrive. Long lines wrap.
- Check for updates, in the command palette and in About Cuewire, checks at once instead of waiting up to six hours.
0.1.7
- In the canvas menu, New agent here opens a list of the agents you can add, and New role… opens the form for a new role.
- A Claude Code agent restarted before you talked to it starts a new session instead of failing.
- Agents learn that notes are not files, and how to write to them.
0.1.6
- When Claude Code starts a sub-agent named after an agent wired to it, Cuewire stops it and tells Claude Code to hand the task to that agent instead.
0.1.5
- Every agent starts knowing it is on a canvas, and how to reach the agents wired to it.
- Ctrl+V pastes text in terminals, and a terminal's right-click menu has Copy and Paste.
- A bigger box for writing prompts.
0.1.4
- Add an agent from the canvas's right-click menu or the command palette.
- A role's instructions get a box of several rows.
- Esc or a click outside closes the layer menu and the usage panel.
0.1.3
- Agents' sign-in pages open in your browser, so you can sign in to Claude Code from Cuewire.
- Ctrl+click opens a web address in a terminal, even one that runs over two rows.
0.1.2
- No new features. The first version that installed copies took as an update by themselves.
0.1.1
- Cuewire updates itself. It offers each new version in the sidebar, downloads and checks it, and installs it when you restart.
- About Cuewire shows the version you are running. Open it from the command palette, or click Cuewire at the top of the sidebar.
- Uninstalling no longer flashes a console window.
0.1.0
- The first alpha, for Windows. Run AI coding agents, such as Claude Code and Codex, in real terminals on one canvas.
- See which agents wait for you in the Needs you list, and answer without leaving what you are doing.
- Wire agents together so they hand work to each other, with notes as their shared memory.
- Give each task its own layer, and review the changes before you merge.