New Features
- Full conversation on an agent task card now opens a real conversation. Each turn shows who spoke and when, the agent’s answers are formatted the way they were written, headings, lists, code, and the tools it ran sit quietly beside them. What the agent is typing right now appears at the bottom in full instead of a single clipped line.
- A forwarded message now reads as a message inside your message. It sits in its own card, headed by the word forwarded and the name of whoever wrote it, so a relayed message is never mistaken for something the sender typed. Open the original moved into that card, next to what it opens.
- A Steer button on an agent task card. Replying to a card has always redirected the work an agent is doing, but that was only reachable by hovering the card; the button says so plainly and puts the card in your reply box in one press.
- A working agent now shows what it has done, not only what it is doing. Its card keeps a short trail of recent steps, the files it read, the searches it ran, the commands it tried, with the current one at the bottom, so it is clear what is happening rather than a single line that keeps changing.
- Every agent task card has a See full conversation view: the agent’s answers, the tools it ran and the steers it was given, live while it works. Conversations are kept for 24 hours after the last activity and then disappear.
- When an agent opens follow-up work, the summary now shows a live row per task, its colour and status update as agents pick the work up, so it is visible when nobody has.
- A task that grew out of earlier work now says so on its card.
- Messages can now be forwarded. Right-click a message, or use the Forward button on its hover bar, and pick any number of channels and direct messages to send it to. It arrives exactly as it was written, pictures and files included, which makes it easy to hand something straight to a bot or agent channel.
- A channel’s agents can be given a preferred model, for the whole channel and per project. Tasks that name no model run on the preference; naming one in the message still wins, and a machine that cannot run it picks its own and says so on the card.
Bug Fixes
- A task with no pull request no longer shows a View pull request button that does nothing.
- The activity dot on an agent card sits on the line it points at instead of below it.
- A conversation that has not been stored yet no longer says nothing was recorded while the agent is visibly working, and a conversation that fails to open now says why.
- A task that finished while you were looking at it no longer keeps its Stop button.
- Opening a channel while an agent is working now shows what it is doing. The card was blank until the agent next changed what it said, which during a long step could be minutes.
- Full conversation is now a button on the task card alongside the others, instead of a small link.
- A merged pull request no longer adds a second card to the channel. The outcome is now written onto the card of the task that opened it, which previously stayed at "pull request open" for ever.
- An agent watching a pull request only posts a card when it has something to report, a failing build, a fix, a review. A pull request that passes and merges cleanly is now silent.
- Connecting an AI client to the local MCP server without a token now explains exactly how to authenticate, instead of failing with a confusing OAuth error.
- The project name field now explains itself when it adjusts what you type: names are lowercase chat tags, while a project’s folders keep their capitalization.
- Deleted groups no longer appear when you right-click a member and open Groups. They used to stay in the list looking exactly like a live group, and assigning one silently did nothing.
- Assigning a group to a member now says so when it cannot be done, instead of leaving you to wonder whether the click registered.
- Deleting or restoring a group takes effect everywhere immediately, including the member list, name colours and the channel pickers.