Modes

Each mode is a different shape of AI-agent session against your repository, from one agent working alone to a planner, developer, and reviewer relay.

Every mode launches its own session. The agent counts below are the ones in the FAQ mode table.

Team mode

The three-agent relay for issue-backed delivery in the current repository: a planner, a developer, and an independent reviewer, each in its own pane. Only the planner starts working on launch; the developer and reviewer panes start idle and wake as work reaches them.

Use it when a multi-file feature should be planned, built, and reviewed by separate agents before it merges.

Agents: 3 (planner, developer, reviewer).

Duo mode

The leaner two-agent delivery loop: a developer who plans and implements, and an independent reviewer who gates both the plan and the implementation. It runs issue-backed delivery in the current repository and keeps a stable worktree per backend combination.

Use it when you want an independent review gate but do not need a separate planner.

Agents: 2 (developer, reviewer).

Adhoc mode

The single-agent delivery path: one agent claims a ticket, plans, implements, opens the pull request, and merges it. With auto-start it selects the next ready ticket itself.

Use it when a small, well-scoped issue can go from claim to merge in one pass.

Agents: 1.

Explore mode

The open-ended investigation mode: it runs with or without a repository, in its own dedicated home, and hands its findings off as tickets rather than code. A finding is reported as confirmed, not confirmed, or insufficient evidence, with a next action.

Use it when you have a question about a codebase or a bug report to reproduce, and you want the evidence filed as a ticket.

Agents: 1.

Audit mode

A per-repository session in a dedicated worktree: each pass refreshes to the latest merged code and reviews it for quality problems, filing tickets but never editing code, opening branches, or merging. It runs on a cadence poller and also watches the continuous-integration surface.

Use it when you want a standing review of already-merged work that turns what it finds into triage tickets.

Agents: 1.

Caucus mode

A two-agent deliberation: a proposer and a challenger converge one topic into a concrete conclusion, rendered as GitHub tickets, with no delivery code. It launches idle and cuts a fresh read-only snapshot from the default branch.

Use it when the team needs to settle a design or strategy question as a concrete issue.

Agents: 2 (proposer, challenger).

Live mode

A hands-on single-agent session with no issue queue, no reviewer, and no relay. It can work directly in the current directory and branch, or shift to patch delivery with a branch, a commit, and a pull request when the change calls for it.

Use it when you want to work interactively with one agent on operational work or a change you are driving yourself.

Agents: 1.

Which mode do I start with?

Start with adhoc: it is the single-agent loop that claims one issue, implements it, opens the pull request, and merges it — the smallest complete loop for seeing the handoff work end to end. That is the mode the Linux quickstart launches in its first-session step. Move up to duo or team when you want an independent review gate, and to explore, audit, or caucus when the work should produce tickets instead of code.