How It Works
Twill lets you choose from multiple coding agents to handle your tasks:- Claude Code — Anthropic’s official CLI with Claude Opus, Sonnet, and Haiku models.
- OpenCode — Open-source agent supporting Claude, GPT, and Gemini models.
- Codex — OpenAI’s coding CLI. Coming soon.
Per-Task Agent Selection
When creating a task, you can select which agents to use:- Choose your agent: Pick from Claude Code, OpenCode, or Codex
- Run in parallel: Select multiple agents to compare their outputs
- Run multiple times: Run the same agent n times to increase success rate
Workspace Settings
Configure defaults from Settings → Agent:| Setting | What It Does |
|---|---|
| Default Agent | Agent selected by default when creating new tasks |
| Plan Approval | Control whether plans need manual approval before implementation |
Plan Approval
Control when plans skip manual review:| Strategy | Behavior |
|---|---|
| Always require | You must approve every plan before implementation |
| Auto-approve | Plans marked as ready are automatically approved |