Claude Code desktop gives each parallel task its own session
If you ask Claude in the desktop app to run tasks in parallel, it now starts a nested session for each one. Each session reports its result back into your main thread when it finishes, so the conversation you started stays readable, as Anthropic explained in a post from the ClaudeDevs account on X.
At a glance
- The Claude Code desktop app gets a new session screen that keeps recent folders one click away, plus saved environments covering several folders, an SSH host or the cloud.
- Parallel tasks run as nested sessions that report back, suggested follow-ups launch as separate sessions in one click, and /btw opens side questions in a tab while work continues.
- The post gives no cap on concurrent sessions, says nothing on how sessions count toward usage limits, and does not say whether a nested session gets its own worktree.
If you have not been following, this is not the first round of work on the desktop app. According to a guide from Miraflow, Anthropic redesigned it on April 14, 2026, with a multi-session sidebar, a drag-and-drop layout, an integrated terminal and file editor, and Git worktree isolation. Before that, the same guide says, parallel work meant separate terminal windows with no unified view across sessions.
Saved environments start a session in several folders, over SSH or in the cloud
Anthropic's pitch has three parts: sessions are now much quicker to start, work is easier to split across a few of them, and you can always see what each one is doing. The starting point is a new session screen that keeps your recent folders one click away, which is how the post frames picking up where you left off.
The bigger addition is environments. You can save a setup, such as a few folders, an SSH host or the cloud, and start new sessions inside it, with the option of a worktree per folder. The desktop quickstart in the Claude Code documentation covers the first run if you have mostly used the CLI so far.
Parallel tasks run as nested sessions that report back to the main thread
When you ask Claude to run tasks in parallel, it starts a nested session for each one. Each nested session reports its result back into your main session when it is done, which, according to Anthropic, keeps the main thread readable even when several tasks are running at the same time.
Claude also suggests follow-up work while it is still busy, and you can run any suggestion as its own session with one click. Side questions work in a similar way through /btw: the question opens in its own tab, and the main session keeps going in the meantime.
That gives you three ways to branch work off the main conversation: delegated parallel tasks, follow-ups Claude suggests, and your own asides. Each one ends up in its own session or tab rather than in the thread you are steering.
The sidebar shows which sessions are working and which need you
With several sessions open, the sidebar shows which ones are busy and which are waiting for your input. Inside a session, a live to-do list follows Claude's plan as it works through the steps, and it updates as the plan moves forward.
When Claude needs you, a recap card sums up the turn. A side rail shows where the session runs, its context and its changes, which helps once some sessions live on an SSH host or in the cloud and others run on your own machine.
Independent tools got here first. According to its GitHub page, Crystal, an Electron app from Stravu, runs several Claude Code instances at once, each in an isolated git worktree, and sends desktop alerts when a session needs input.
A worktree gives each session its own files and branch
The Claude Code docs define a git worktree as a separate working directory with its own files and branch, sharing the repository's history and remote with the main checkout. Without one, two sessions on the same checkout edit the same working tree, so one agent's refactor lands in the other's diff.
Picture one pantry and several cutting boards. Everyone cooks from the same stock, but nobody chops on someone else's board, and whatever one cook leaves on the counter stays out of the next cook's dish.
Git lets a branch be checked out in only one worktree at a time, and a fresh checkout means gitignored files such as .env do not carry over; a .worktreeinclude file copies them in automatically. According to the Claude Code docs, the CLI flag --worktree creates worktrees under .claude/worktrees/ on a new branch, and from v2.1.206 Claude asks for approval before entering a worktree path outside that folder.
Oddly, the post is most specific about how sessions look and least specific about where they write. Worktrees appear only as a per-folder option for environments, and the Claude Code docs describe worktrees, subagents and cross-session messaging as separate ways to run Claude in parallel, so a nested session may share its parent's checkout. The post does not settle it either way.
Session caps and usage still unannounced
Getting the features takes no waiting: update the desktop app and start a new session. What the announcement leaves open is how many sessions can run at once, how they count toward usage limits, and whether a parallel task gets a worktree of its own. Until one of those is answered, check the side rail's list of each session's changes before you merge anything.
Related stories
- Leak puts expandable usage limits in Claude Code desktop
- Leaked Claude Code Hub lists local and cloud sessions
- Claude Code Projects admits its whole Pro and Max waitlist
- In Claude Code, a "block" hook could let commands through
- Claude Code Projects threads can now run on your machine
- Claude Code 2.1.251 streams subagent work live
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.
