- Several attempts, one review
- Send one task to two agents or models, then Review attempts… lays them side by side from one baseline: files, commits, CI and the checks each agent says it ran.
- Orchestrator chats
- One chat starts Claude Code and Codex chats, messages them, waits for them and reads what they changed. Every chat it starts is one you can open and answer.
- Watch and wake
- A chat follows its pull or merge request and wakes its agent when a check fails, a reviewer comments or the branch starts to conflict.
- Limits you can see
- Each login's 5-hour and weekly limits with their reset time and tokens by model. A new chat starts on the login with the most room, and a chat a limit stopped resumes at the reset.
- Teams of agents
- Start a team or a loop from a template, such as Write and cross-review, or your own. A coordinator hands work items to roles, each with its own agent, login, model and worktree, and accepts one only after its check passes. A loop passes each reply to the other chat for the rounds you set.
- Context from your review
- Attach a diff hunk, a file, a region of an image marked in its own pixels, or a golden-test failure with its expected, actual and difference images. The agent reads the text and sees the images.
- On a schedule or from an issue
- Run a task template at a set time on weekdays or every day while ffgit runs, or start a chat from a GitHub or GitLab issue, quoted into the task as data.
- Live status, one inbox
- Working, waiting, ready or failed, as each agent reports it, on the sidebar and the board; Waiting for you (⌘⇧I) lists the chats that wait on you. Questions, ready responses and failures land in Activity.
- Hand off to another agent
- Continue in Codex or Continue in Claude Code opens a new chat for the other agent with the first request, the latest messages and command outcomes, for you to read, edit and send.