Skip to content

Run agents

Superconductor runs local coding agents. It gives each agent an isolated Git worktree, a terminal, a chat surface, and review tools. Prompts never pass through Superconductor servers.

Agents run with the provider tools and credentials you already use: Claude Code, Codex, Pi, Grok, Cursor Agent, Gemini CLI, Kiro, OpenCode, Copilot, or another CLI-based tool.

Before launching an agent in Superconductor, confirm its command works in your normal terminal.

  1. Open the workspace for the project.
  2. Select or create the worktree for the task.
  3. Confirm the terminal sits in the correct project or worktree directory.
  4. Start your agent CLI or open a configured chat session.
  5. Give the agent its task.

Create a separate worktree for each Git-backed task. Worktrees isolate file changes, terminal state, branches, and review surfaces while keeping related work inside the same durable workspace.

Review surfaces show the diff, the changed files, and the task context for the selected worktree.

Superconductor never sends your code or prompts through Superconductor servers. Agent traffic flows through the configured agent CLI or native provider adapter.