Claude Code in Superset

Overview

Claude Code is Superset's most deeply integrated agent. Each session runs in its own workspace, with its own Git worktree and terminal, on your Claude plan or API key. Superset never proxies model calls.

Launch

Superset starts Claude Code with:

claude --dangerously-skip-permissions

This turns off Claude Code's approval prompts, so the agent can edit files and run commands in its worktree without asking. To keep the prompts, open Settings → Agents → Claude and remove the flag from Command and Resume args. Worktrees separate files, but do not sandbox processes; see Workspaces.

What Superset Adds

• Status and notifications - The sidebar shows when Claude Code is working, needs input, or is done, and you get a notification. See Agent Status.
• Resume and fork - Superset restores a session after a restart with --resume, and can fork a session with --fork-session. See Agent Sessions.
• Subagents - Subagents appear under their parent session in the terminal header's Subagents menu.
• Handoff - Moving a conversation to another agent carries Claude Code's structured transcript.
• Model and effort - Pick a Claude model and a reasoning effort when you launch. See AI Agents.
• Usage and accounts - The Usage page shows quota meters and token costs, and lets you add several Claude accounts and choose the default.

Run Several Sessions

Press ⌘N for each task. Each workspace gets its own worktree, so sessions never edit the same files. To give the same task to Claude Code and another agent and keep the best result, see Race Agents on One Task.