Arena Mode

Arena Mode

Arena Mode runs the same prompt through multiple AI providers simultaneously and presents the results side-by-side so you can pick the best answer or merge outputs.

How it works

1. You send a prompt in Arena Mode (toggle in session settings or via clawd session --arena)
2. The daemon dispatches the request to 2-4 configured providers in parallel
3. Each response streams into a separate panel in the desktop app
4. You choose a winner, merge hunks, or discard all and retry

Use cases

• Compare code quality between Claude, Codex, and GPT-4 on the same refactor
• Evaluate model regression when switching provider versions
• Get consensus on architectural decisions — agree = confident, disagree = needs human input

Status

Planned for Sprint GG. Requires multi-session dispatch in the daemon + arena layout in the desktop app.

Related

• Providers
• Session Manager
• Model Intelligence