Skip to content

Multi-Agent Support

CACD supports multiple AI coding assistants. You can run different agents side-by-side, each on its own worktree.

During setup, CACD checks for these agents and adds any it finds:

Anthropic’s CLI coding assistant.

Command: claude

Options:

  • YOLO Mode - Skip permission prompts (--dangerously-skip-permissions)
  • Continue - Continue the last conversation (--continue)
  • Resume - Resume a specific conversation (--resume)
  • Model - Choose between Sonnet, Opus, or Haiku

Google’s AI assistant for the command line.

Command: gemini

Options:

  • Model - Select the Gemini model to use

OpenAI’s coding assistant.

Command: codex

Options:

  • Model - Select the model to use

The CLI interface for Cursor’s AI capabilities.

Command: cursor agent

Command: droid

Command: kilocode

Command: opencode

A plain shell session without any AI. Always available regardless of detection.

Command: $SHELL (your default shell)

These agents have built-in state detection patterns but aren’t auto-detected. You can add them as custom agent profiles:

  • GitHub Copilot CLI (gh copilot)
  • Cline

CACD monitors each agent’s terminal output to detect its state (idle, busy, waiting). Each agent has its own detection patterns since they display prompts and status differently.

This means CACD can tell you when an agent needs your attention, even if you’re looking at a different session.

You can create custom agent profiles for any CLI tool. See Agent Profiles for details on setting up your own configurations.

Each session is tied to one agent, but you can:

  • Run multiple sessions with different agents simultaneously
  • Create a new session with a different agent on the same or different worktree
  • Configure which agent is your default for new sessions