Parallel agents in Claude Code: the desktop redesign

Claude Code's desktop app was redesigned in April 2026 to support running agents across multiple projects simultaneously. The previous design assumed you were working in one place at a time. This one assumes you are routing work across several repos and switching as results come in.

Here's what changed.

The sidebar

The main change is a session sidebar that replaces the previous single-session view.

You can filter sessions by status (running, waiting, completed), by project, or by environment. Sessions can be grouped by project — useful when you have multiple branches of the same codebase in flight. When a session's pull request merges or closes on GitHub, the session automatically archives and leaves the active list.

This means you can kick off work on three different repos — a bug fix, a feature branch, and a dependency update — and move between them as each one needs attention instead of waiting for one to complete before starting the next.

Side chat

A new side chat feature (⌘ + ; on macOS, Ctrl + ; on Windows/Linux) lets you branch a conversation without pausing the main task.

This is useful for asking questions mid-session or exploring an alternative approach while the primary agent is still running. The side chat runs as a separate conversation in the same project context; it does not interrupt the main session.

Integrated tools

Three tools that previously required switching out of the app are now built in:

Terminal. Run test suites, build commands, or any shell command directly inside the app. The terminal is attached to the current session's working directory.

File editor. For spot edits — fixing a typo the agent missed, tweaking a config file — without opening your external editor.

Diff viewer. Rebuilt specifically for large changesets. Previous versions of the diff viewer struggled with PRs involving hundreds of files. The new one handles them.

Preview pane. Renders HTML files, PDFs, and locally running app servers. For frontend work, you can preview the running app in the same window where the agent is building it.

All panes are drag-and-drop. Resize and reorder them into whatever layout works for the task.

View modes

Three verbosity levels control what you see during a session:

• Verbose: Every tool call, every file read, every decision the agent makes
• Normal: Tool calls and significant actions, without the low-level detail
• Summary: The agent's final status updates only, hiding most of the working trace

Switch modes mid-session with Ctrl+O. If you are managing multiple sessions at once, Summary mode across the background ones and Verbose on the active one is a practical starting point.

SSH support on macOS

The desktop app now supports SSH for macOS in addition to the existing Linux support. You can connect to a remote server and run Claude Code sessions there directly from the desktop app.

Organizational plugins

For enterprise teams, centrally managed plugins work the same way in the desktop app as in terminal sessions. Administrators can push plugin configurations that apply to the desktop app without requiring each developer to set them up individually.

What to expect workflow-wise

The practical shift is from sequential to parallel. Instead of starting a task, waiting, reviewing, starting the next one — you queue several things at once and review as they complete.

The sidebar and session filtering are designed around that cycle. The most useful pattern is probably: kick off three or four tasks at the start of a work session, switch to something else, and return to the Claude Code app when a session has output to review.

Side chat handles the micro-interruptions — questions and small diversions — without forcing you to interrupt a running session.

Getting the update: The desktop app updates automatically. Version 2.1.111 includes Opus 4.7 support and the new xhigh effort level alongside this redesign.

Official blog post: claude.com/blog/claude-code-desktop-redesign