Skip to content
OpenChamber
↑↓Navigate ↵Open escClose

Sessions in Work

Every session you send a message in puts its work into the In work block at the top of the sidebar, and it stays there until you mark it done, however long ago you last opened it. Your projects keep everything you’ve ever done; In work holds only what you haven’t finished.

What a work is

The sidebar lists works, not single sessions. A work is either:

  • a worktree with every session in it, shown as one row with the task’s name (see Worktree Sessions), or
  • a session in your project’s own folder, with what grew out of it nested underneath: its subagents, its review, its forks and side questions you kept, and sessions its agent started for you.

Click the row to open it; the arrow on its left opens the sessions under it.

Where to find it

The In work block sits in the sidebar under Chats, above your projects or the list by time. A work in it leaves its project, so each work shows once. To keep it under its project too, turn on Also show in work under projects in the sidebar’s display menu. Search lists it once, in the In work block.

Each row takes one line: the title and its status, with a coloured mark when it has a pull request or issue. Rows are ordered by latest activity, so they show no date. Hover a row to see its project, branch, time and links, or turn on Detailed rows in the display menu to show them under the title. Detailed rows apply under your projects too; the sessions inside a work stay on one line.

Pinned sessions have their own Pinned block above In work; to keep them only under their projects, turn off Show pinned section in the same menu. A pinned session in work is listed in Pinned only.

The phone’s sessions list is the same, with the same display menu. Its rows take two lines unless you turn off Detailed rows, and the arrow that opens a work is always there, since there is nothing to hover.

How sessions get in

Send a message in a session and it goes into In work. This covers any project session, including sessions in worktrees. A new worktree goes in as soon as you create it, while the draft that opens in it is still unsent.

A session you marked done comes back when you send a new message in it.

Chats, archived sessions, review sessions and runs of scheduled tasks never go in on their own; fork a run to keep working on it, and the fork does. A subagent never puts its parent in work.

Task names

The agent names each work after its task as soon as the task is clear, in a few words such as “Login timeout fix”. A worktree keeps the name in its branch description, so you see it everywhere you open the repository; a session in the project folder takes the name as its title. Rename it yourself and your name stays: the agent never renames a task you named.

Track and finish a session by hand

  • Track: hover a row and press the eye, or choose Track as in work in the session menu.
  • Finish: hover a row in the block and press the check, or choose Mark done in the menu. Done finishes the whole work: every session in it that was still in work.

Done on a worktree takes it out of In work and leaves it where it is. To also archive it when you mark it done, turn on Archive a worktree when you mark it done in Settings → Sessions: the folder goes, and its sessions, changes and branch wait on the Archive page, ready to restore. If a session there is still running, the worktree stays.

On the phone, swipe a row to the side to reach the same action.

Chats don’t get the eye: they’re plain conversations.

”This work looks done”

With a classification provider set up, Jev checks each turn of a session in work. When a turn looks like the end of the work (you said thanks, asked for a commit, confirmed it works), Jev leaves a hint instead of closing anything:

  • a grey check on the session row in the sidebar
  • a strip above the composer: This work looks done. with a Mark done button

Press Mark done if Jev is right. If not, just keep going: the hint disappears with your next message. Without a classification provider there’s no hint, and you mark sessions done yourself.

Recap and next-message suggestion

After the agent finishes and the session stays quiet for a minute, OpenChamber can write two things with your Small Model (see Providers, Models & Agents):

  • Recap: a short reminder of what was done and where things stand, shown under the last reply and in the session’s tooltip in the sidebar
  • Suggestion: the next message you’d probably send, as a row at the top of the composer; click it to put the text in the composer, it never sends by itself

With Jev available, OpenChamber first asks it whether the turn needs them. A plain “thanks, that’s it” after real work still gets a recap, but a suggestion is skipped when nothing is left to do, and the Small Model isn’t called at all when neither is useful. Without Jev both are generated after every turn, as before.

Turn them on or off in Settings → Chat → Session Assistance.

Settings

Open Settings → Sessions → Sessions in work. Show sessions in work is on by default. Turn it off and the whole feature is off: no block, and sending a message no longer puts sessions in work.

To hide only the block, use Show in work section in the sidebar’s display menu.

Session cleanup keeps sessions in the In work block. To let them expire like any other session, turn on Include sessions in In work under Settings → Sessions → Session retention. It’s off by default, and pinned sessions stay either way.

Where it works

The block and the manual track/done work everywhere, VS Code included. Tracking sessions when you send a message, the “looks done” hint and the Jev filter for recaps need the OpenChamber server, so they work in the desktop app, the web app and on your phone, but not in VS Code. In VS Code you track sessions by hand.