The app runs the coding-agent tools and accounts you already use. Each session starts in the selected worktree, so its code changes, terminal state, chat history, diff, and review context stay attached to the task.
Choose how the agent runs
Section titled “Choose how the agent runs”| Surface | Choose it when |
|---|---|
| Chat UI | You want rendered markdown, inline diffs, attachments, and app-native review handoff |
| Terminal | You want the provider’s complete native CLI and its latest provider-specific behavior |
Provider traffic goes directly from the local provider process to that provider. The app does not proxy prompts or store provider credentials.
Start with the smallest useful setup
Section titled “Start with the smallest useful setup”- Confirm the provider CLI works in your normal terminal.
- Select or create the worktree that owns the task.
- Open a Chat UI or Terminal session.
- State the outcome, relevant constraints, and how to verify the result.
- Inspect the resulting diff before taking the next Git action.
Use separate worktrees for independent branch-backed tasks. Use several sessions in one worktree only when they intentionally share the same branch and ownership model.
Know when to orchestrate
Section titled “Know when to orchestrate”A single agent is the clearest choice for one bounded task. Use orchestration when the work benefits from explicit roles, cross-provider collaboration, direct agent communication, staged handoffs, or independent review. The user defines the outcome and constraints; the lead can choose the smallest useful team after inspecting the repository.
Choose your next topic
Section titled “Choose your next topic”- Configure available tools in Providers and models.
- Control launches with Custom commands and routing.
- Follow the single-agent workflow in Run agents.
- Attach the right material with Files, mentions, and context.
- Control interruptions in Notifications and approvals.
- Return to previous work with Session history and restore.