Anthropic rebuilt Claude Code desktop from scratch with parallel sessions, integrated terminal, and file editor
Side chat (Cmd+;) branches questions off a running task without polluting the main session
Three view modes (Verbose, Normal, Summary) let you control exactly how much of Claude's work you see
Claude Code on the web and iOS means your coding agent travels with you
The preview pane runs local servers in-app, eliminating the tab-switching development loop
On April 14, 2026, Anthropic shipped the biggest update to Claude Code since the tool launched. Not a feature addition. Not a UI tweak. A complete architectural rebuild of the desktop application. The new Claude Code desktop is built around parallel sessions, an integrated development environment, and a workflow design that treats the AI agent as a first-class citizen rather than a chat window bolted onto a terminal.
I have been using Claude Code daily since the CLI launched. The desktop rebuild changes how I work with it.
Parallel Sessions Change the Game
The old Claude Code desktop ran one session at a time. You could have multiple windows open, but each was an isolated conversation with no unified management. The redesign replaces this with a sidebar that manages all active sessions from a single window.
This sounds like a minor UI improvement. It is not. Parallel sessions mean you can have Claude working on a frontend component in one session, debugging an API endpoint in another, and writing tests in a third. All visible at once. All running concurrently. All managed from the same interface.
The sidebar supports filtering by status, project, and environment. When you are juggling five sessions across two repos, this filtering is the difference between organized workflow and chaos. I typically run two to three sessions in parallel during a development sprint: one for the main implementation, one for edge case testing, and one as a research thread where I ask Claude questions about the codebase without contaminating the implementation context.
The drag-and-drop layout lets you arrange sessions however you want. Side by side. Stacked. One maximized with the others minimized. The layout persists between restarts, so your workspace configuration survives closing your laptop at the end of the day.
The Integrated Terminal Is Not Optional
Previous versions of Claude Code desktop relied on your external terminal. The AI would suggest commands, and you would run them in a separate window. The rebuild embeds a full terminal directly into the application.
This changes the feedback loop dramatically. When Claude suggests running a test suite, you see the output in the same window without alt-tabbing. When a build fails, the error output appears right next to Claude's analysis of what went wrong. The context switching that used to fragment every debugging session is gone.
The terminal is not just a display pane. It is a fully functional terminal emulator. You can run arbitrary commands, manage git operations, start development servers, and monitor logs. Everything that previously required a separate iTerm or Terminal window now lives inside Claude Code.
For my workflow, this collapsed three application windows (Claude Code, terminal, editor) into one. I still use VS Code for complex refactoring sessions, but for the majority of Claude-assisted development, the integrated terminal covers everything.
Side Chat Solves the Context Problem
This is the feature I did not know I needed. Side chat, triggered with Cmd+; on Mac, opens a secondary conversation that branches off the current session. The branch shares the same project context but does not feed its conversation back into the main session.
The use case is immediate. You are in the middle of a complex implementation session. Claude has built up significant context about what you are building and why. You want to ask a quick question: "What is the return type of this function?" or "Does this package support ESM?" In the old model, asking that question polluted the main session with tangential context. The AI would factor that question into its subsequent responses, sometimes drifting off course.
Side chat isolates these tangential questions. Ask whatever you need. Get the answer. Close the side chat. The main session never sees it. Claude's understanding of your implementation task remains clean and focused.
I use side chat constantly now. Quick API lookups. Package compatibility questions. "How does this existing function work?" queries that inform my next instruction to the main session. It is like having a reference desk next to your workbench.
Three View Modes for Different Contexts
The redesign introduces three view modes: Verbose, Normal, and Summary.
Verbose shows everything. Every tool call, every file read, every command execution, every reasoning step. This is useful when you need to understand exactly what Claude is doing, either for debugging or for learning from its approach.
Normal is the default. It shows key actions and results without the granular detail. You see that Claude edited a file and what changed, but not the intermediate steps it took to decide what to change.
Summary is the mode I use most. It shows only the outcomes: what files were created or modified, what commands were run and whether they succeeded, and any questions Claude has for you. Everything else is collapsed. When Claude is working on a task you have already aligned on, Summary mode lets you monitor progress without drowning in detail.
The ability to switch modes mid-session is what makes this useful. I typically start a new task in Normal mode while I am actively collaborating with Claude, then switch to Summary once the plan is clear and Claude is executing. If something goes wrong, I switch to Verbose to diagnose.
SSH Support on Mac (Finally)
Remote development sessions now work natively on Mac. You can connect Claude Code to a remote server via SSH and work on a remote codebase with the same capabilities as local development.
This was possible before through terminal workarounds, but native support means Claude Code handles connection management, reconnection on network interruptions, and file synchronization. For anyone working with remote development environments (staging servers, cloud VMs, GPU instances), this removes a significant friction point.
Claude Code on the Web and iOS
The desktop rebuild coincides with Claude Code becoming available in a browser at claude.ai/code. Connect a GitHub repo, describe a task, and Claude works in an isolated cloud environment. Multiple tasks run in parallel across different repos, each producing its own pull request.
The web version is not a watered-down experience. It has real-time progress indicators, file browsing, diff views, and the same model powering the desktop app. The limitation is that it runs in Anthropic's cloud rather than your local machine, so tasks that require local dependencies or private infrastructure still need the desktop app.
The iOS experience is the unexpected entry. You can start a coding task from your phone, let Claude work on it while you are away from your desk, and review the results later. I have used this exactly once (fixing a typo in production while on the U-Bahn) and it worked. Not a daily driver, but knowing the option exists changes how I think about response time for urgent fixes.
What This Actually Means for Daily Work
The desktop rebuild is not about any single feature. It is about the interaction model.
Before this update, Claude Code was a very capable tool that you used through a chat interface. You told it what to do. It did it. You reviewed the result. The new desktop app treats Claude Code as an environment rather than a tool. You do not launch it to accomplish a specific task and then close it. You open it at the start of your workday and it stays open, managing multiple workstreams, surfacing progress, and waiting for the next instruction.
The parallel sessions, integrated terminal, file editor, and preview pane combine to create something that looks a lot like an IDE. But unlike a traditional IDE where the developer writes every line, this IDE has a collaborator built in. One that reads the whole codebase, remembers the conversation history, and executes multi-file changes autonomously.
For the solo developer running a one-person studio, this changes the math on what is achievable in a day. I used to cap my active projects at two or three because context switching between codebases took real time. With parallel sessions, each maintaining its own project context, I can move between projects without losing state. Claude remembers where each project left off. I just need to decide which one to push forward next.
The rebuild is available now for Pro, Max, Team, and Enterprise plans. If you have been using the CLI exclusively, try the desktop app this week. The parallel session management alone is worth the switch. The integrated terminal and side chat make it hard to go back.
Top comments (0)