For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt, and this page is available as Markdown at /changelog/2026-09-24.md.
More than one chat with the same Agent on a task: mentioning an Agent or Crew in a comment now starts a new chat on the task, attached to that comment. Several chats with the same Agent can coexist on one task without replacing each other. Start another one from New chat in the conversation panel or Start a new chat with an Agent in the Agent menu; the roster shows how many chats already exist. The Assignee tag marks only the originally assigned chat.
Automation workflows from the CLI: browse templates, create a draft, deploy, run, and read run history from the terminal.
Quitting the desktop app no longer stops the local service: quitting now shows a confirmation and keeps the CLI daemon running in the background by default, so Agents keep taking work; stopping the service is an explicit choice. Closing the main window just hides it to the tray instead of needing two attempts to quit.
Open an Automation from the Inbox: selecting an Automation run notification opens the Automation's full details in the right-side drawer instead of leaving the Inbox. Closing the drawer returns you to the list.
Easier to read run output: a finished turn collapses its process and shows how long it took, while a turn that ended abnormally stays expanded so you can see how far it got.
Conversations open at the newest message: opening a chat no longer starts at the top, and earlier content loads as you scroll.
Better global search: the all-types entry is always available, a selected type becomes a chip in the search box that you can exit while keeping your keyword, results can be filtered by Space, and matches are shown in full with the title, task identifier, and summary keywords highlighted.
The one-time Webhook URL is readable: the panel shown after you create a Webhook trigger has been relaid out — the full URL fits, copies in one click, and the buttons sit in a fixed place instead of floating in an empty card.
Agent instructions are now WYSIWYG: the instructions field in Agent settings is no longer a syntax-highlighted Markdown source box; what you type is what you see, and it stays readable in dark mode.
Setting a default working directory is smoother: reopening an already-inspected default working directory shows the isolated/direct choice immediately instead of waiting for the check, and it now says so when the folder is gone, is no longer a Git repository, or can no longer use isolated worktrees.
An Agent mentioning another Agent in a comment now triggers it: previously the mentioned Agent did not respond. If that Agent already has a pending run on the task, the mention is skipped so Agents do not trigger each other in a loop.
One-click repair fixes the local service: a failed graceful stop now falls back to force-terminating and reclaiming, so the repair no longer aborts. Restarting the service manually is immediate and no longer waits for running tasks.
Runtime probing no longer leaves empty conversations: probing MiMo, OpenCode, CodeBuddy, and similar Runtimes no longer adds "Untitled" conversations to your list, and no longer triggers an authorization prompt.
Windows computer ownership transfer can recover: retrying the local recovery no longer fails every time, so you no longer have to cancel and reconnect.
Search no longer misses tasks: searching no longer drops tasks that clearly exist.
An Agent's thinking depth is no longer dropped: a thinking depth set through Custom arguments now reaches the Runtime, and when the Runtime cannot apply it before the first prompt the run stops with an error instead of silently using the default. A thinking depth picked on the Agent's model row is applied to both new and resumed chats.
New users no longer see the sign-in page flash: finishing Space setup goes straight to the signed-in state.
Manage statuses no longer leaves an unclosable dialog: opening status management from the create-task dialog no longer leaves the original dialog stuck on top.
Tasks without Customers hide the empty Customers block: the block is not rendered when a task has no linked Customer Request; create one from New request from next to the subtask action under the description.
Agents can create Automations for you: asking an Agent in a chat to create an Automation no longer fails with a permission error (viewing and creating are allowed; editing and deleting still require a person).
