Anthropic rebuilt the Claude Code desktop app and added several features. These include projects, threads, side chats, split view, mods, and new slash commands. Claude Code runs on the same underlying tool as Claude Cowork. It offers more functions, including access to any file on the user's computer and to command-line interfaces (CLIs).

Layout Upgrades

The side panel now shows the same information as the Cowork side panel, plus additional elements:

• To-do list: When Claude creates a plan, the chat displays a to-do list. The user can track each task as Claude completes it.
• Environment card: Shows the folder the chat runs in and whether the chat is local or cloud-based.
• Remote control switch: When enabled, lets the user continue a local chat in the cloud, for example from a phone.
• Context card: Shows how full the chat's context window is and what occupies it. A shortcut button compacts the chat. Compaction summarizes the conversation to free space when it reaches a set number of tokens, which helps maintain performance.
• Connectors card and outputs card: Show the connectors used in the chat and every file the chat created or used.
• Chat sections: Users can group chats by category. To create a section, click the three dots next to a chat, select "move to new section," name it, and create it. Chats can then be moved into the section by drag and drop. The previous grouping feature lacked this function.

Projects and Nested Sessions

Projects are available on Pro and Max plans. Team plans receive nested sessions instead. Anthropic is rolling out both features gradually, so availability may differ by account. Both work on the same principle.

A main chat, or project, can start other chats, called nested sessions or threads. These sessions can send messages, check their own work, and report back. A single coordinator, or orchestrator, manages a small group of sessions. The user works with one chat instead of managing several.

Example: preparing a YouTube video

The presenter describes three separate tasks in one production workflow:

1. Thumbnails and titles
2. Outline
3. Visuals and editing

Before this update, the separate chats had no shared context. A change in one chat did not reach the others. Working in a single chat caused the context window to fill quickly, which reduced performance. With nested sessions, the coordinator creates one thread per task. The user can iterate on each thread from the main chat, and the coordinator tracks progress across all threads.

Threads can spawn their own sub-agents. Users cannot iterate directly with sub-agents, but they can iterate with threads and nested sessions. The coordinator therefore adds one layer of agents above existing sub-agents.

Access and setup

• With access to projects, starting a new project automatically creates threads.
• Team plan users without project access can open a blank chat and ask Claude to work on a larger project. Claude then offers to create threads, or the user can instruct it directly.
• When a user asks an unrelated question mid-chat, Claude offers to start a new thread.
• Created threads appear in a drop-down below the main chat, where the user can continue working with each agent.

Differences between projects and nested sessions

Anthropic plans to make projects available on all plans. The presenter expects this within a couple of weeks.

Side Chats

Side chats allow users to ask questions without interrupting a running task. To open one, click the three dots in the top-right corner and select "side chat." Each side chat opens in its own window, and users can open several.

A side chat can read the main chat up to the point it was opened. Users can ask what the main chat is doing, how long a task may take, or request a review of the work so far. These queries do not add content to the main chat's context window.

Side chats are persistent. They are saved, and they can send messages to the main chat. The presenter uses this feature to have a side chat review the main chat's output. The feature is a more advanced version of the temporary side chat slash command.

Split View

Split view displays several chats on one screen. Users can:

• Drag a chat from the sidebar into the current chat and select the split view option.
• Click the three dots next to a chat in the sidebar, select "open in," and then select "split view."

Docs, Slides, Design, Dashboards, and Motion

Anthropic is building document, presentation, design, dashboard, and video tools inside Claude. Each output is an artifact. Artifacts function as editable canvases that support comments. Users can publish an artifact as a live page with a shareable link.

• Dashboards: Recently launched. Dashboards pull data directly from connectors. Each time a user opens a dashboard, it refreshes with the latest data from those connectors.
• Slides: Creates animated presentations and slideshows. Users can download them in PPT format or present them directly in Claude.
• Docs: Functions as a document editor with tabs. Team members and AI agents can collaborate on documents in real time. The presenter uses Docs to prepare all YouTube videos.
• Design: Provides a Figma-like canvas. Users can change copy, leave comments, and instruct Claude to make corrections. The presenter uses Design for social media and website work.
• Motion: Creates animated videos and motion graphics by generating code. Claude Motion does not use a video generation model. The presenter states that this approach costs significantly less than a video model. Users can iterate on videos on the canvas.

Slash Commands

Claude Code includes more slash commands than Cowork. A slash command is a shortcut that starts with a slash. The update adds four commands worth knowing.

/output-style

Controls the format and tone of Claude's responses. The presenter created a custom style that keeps answers short and simple, because default responses were often too technical. To create a custom style, ask Claude to build one, add it as an output style, and invoke it with the slash command.

/doctor

Reads the CLAUDE.md files, which contain the first instructions Claude reads in every chat. It also reads all skills and settings. The command identifies outdated or conflicting instructions. These may be unnecessary for newer models.

In the presenter's account, the command analyzed 404 files and found 312 problems. Among them were instructions no longer needed for Opus 5.5 that consumed unnecessary tokens.

/skill-doctor

Each installed skill uses part of Claude's context window, even when the skill is not in use. The command reports the token cost of each skill and how often the user actually applies it. It flags unused skills and recommends which ones to delete. The presenter reports that removing unused skills saved about 20,000 tokens.

/advisor

Allows a lower-cost model, such as Sonnet or Haiku, to request advice from a stronger model, such as Opus, at key points in a task. The user receives the stronger model's judgment on critical steps without paying for that model throughout the entire task.

The presenter uses Sonnet to clean an email inbox together with /advisor. When a message requires a critical response, Opus 5.5 helps draft it.

Mods

A mod changes how Claude Code looks and behaves. Users can build features inside the Claude Code desktop app. To create a mod, describe the required function to Claude and ask it to build the mod.

The presenter built a brain dump mod. Voice brain dumps often contain errors, lack structure, and sometimes contradict themselves. The mod detects a long voice brain dump and converts it into a clear, structured prompt before Claude processes it.