Starting a Session

There are several ways to start a new chat:

• Click "New Chat" in the sidebar (or press Cmd+N / Ctrl+N)
• Click a project name in the sidebar
• Click "Open" on a project card in the Dashboard
• Use the Command Palette (Cmd+K) and select "New Chat"

If you have multiple projects, the "New Chat" button becomes a split button — click the arrow to choose which project to chat about.

The Workspace

When a session opens, you see the Workspace with two main view modes:

Chat Mode (Default for SDK backend)

A visual chat interface with:

• A message thread showing your conversation
• A text input at the bottom for sending messages
• Tool call cards showing what Claude is doing
• Plan and question cards for interactive decisions

Terminal Mode (Default for CLI backend)

The full Claude CLI terminal interface running in a PTY. This is the raw Claude Code experience — you see exactly what the CLI shows, including its interactive TUI.

You can switch between modes using the Chat and Terminal tabs at the top of the workspace, or press Cmd+Shift+T / Ctrl+Shift+T.

Sending a Message

1. Type your message in the input field at the bottom
2. Press Enter to send (or click the send button)
3. You can attach images by dragging them onto the chat, or by pasting from clipboard

While Claude is Thinking

When Claude is processing your message:

• The status dot in the header turns blue (animated)
• The input shows the current model and mode
• You can queue messages — type and send while Claude is busy. Messages queue up and send automatically when Claude finishes
• Press Escape to abort the current operation

Message Queue

If you send messages while Claude is thinking, they appear in a queue above the input:

• Each queued message shows a preview and a remove button
• Messages are sent in order once Claude becomes idle
• You can remove or reorder messages before they're sent

Tool Calls

As Claude works, you'll see Tool Call Cards in the chat. These show:

• The tool name (Read, Write, Edit, Bash, Glob, Grep, etc.)
• A running timer showing how long the operation takes
• Input parameters (expandable)
• Output or error messages
• Status indicators: spinning (running), green check (success), red X (error)

Click a tool call card to expand/collapse its details.

Plans

When Claude proposes a plan (for complex tasks), a Plan Card appears:

• Shows the plan title and description with full markdown rendering
• Accept or Reject buttons at the bottom
• If you reject, you can optionally provide a reason

The chat input is replaced with a plan review banner until you respond.

Questions

When Claude asks a question (e.g., choosing between approaches), a Question Card appears:

• Shows the question with multiple choice options
• Select one or more answers
• Click Submit to send your response

Subagents

For complex tasks, Claude may spawn subagents — child agents that work independently. These appear as Subagent Cards:

• Shows the agent type (e.g., "Explore", "general-purpose") with a colored badge
• Description of what the agent is doing
• Running timer and tool use count
• Expandable to see the agent's full activity (messages and tool calls)
• Nested subagents are displayed recursively

Status Indicators

The status dot in the session header shows the current state:

• Gray — Idle, ready for input
• Blue (animated) — Thinking, processing your message
• Amber — Waiting for your input (plan approval, question, permission)
• Red — Error occurred

Next Steps

• Workspace Overview — Learn about all workspace panels
• File Explorer — Browse your project files
• Browser Preview — Preview your running app