Multilo 2.0.174 is out on Mac now; the Windows update follows shortly.
Send while a chat is working
You no longer have to wait for a run to finish, or stop it, to say something.
- Enter steers: while the agent works, the button turns back into Send as soon as you type, and your message joins the run at the agent's next step.
- Tab queues the message for when the run ends.
- Cmd/Ctrl+Enter stops the run and sends your message straight after.
- Stop keeps your queue: anything you queued goes next. Remove a queued message with the ✕ on its chip.
- Your choice of default: in Settings → AI Agent → Messages sent while the agent works, pick Steer or Queue for what Enter does.
A browser that keeps each chat's pages
- Pages stay with their chat. The pages a chat opens show while that chat is on screen. Switch chats or projects and they are put away, not closed, so a running chat keeps working in them and they come back when you return.
- Your own tabs come back with their project, and after a restart.
- Chats only use their own pages. An agent can't read or click your tabs, or another chat's pages, until you turn on Share with Multilo after the address bar. Agents can also close the pages they opened, and deleting a chat closes its pages.
- Local files open in the browser. Type a file path in the address bar, or right-click an
.htmlfile in the explorer and choose Open in Browser. - Cmd/Ctrl+Shift+B shows and hides the browser (View → Toggle Browser), and app shortcuts such as Cmd/Ctrl+P, Cmd/Ctrl+Shift+F and Cmd/Ctrl+` now work while a web page has focus.
Start a chat from a link
A multilo://open?q=… link opens a new chat with the prompt in its box, and cwd= can name the folder. Nothing is sent until you press Enter, and a line under the box says the prompt came from an external link. Course pages and supervisors can now share "start this task" in one click. multilo://settings/privacy and other settings links open Settings at that page.
For organisations
Multilo can export usage metrics and events over OpenTelemetry to your own monitoring, using the same settings as Claude Code. It is off unless an administrator turns it on, and prompts are left out unless they choose to include them.
Also in this update
- Auto explains itself: the model picker now says Auto picks one model for each chat and keeps it.
- Line counts match: a file card's +/− counts now agree with the "files changed" bar.
- Going back to a project shows its last chat again, not a blank new one.