Agents
Agent CLIs
Each turn spawns a local agent CLI. Settings picks the default. The composer can pin a different CLI and model on the open chat.
Choose a harness
Settings → Agent sets the default for new chats. The open chat can pin another CLI and model. Those pins stick for later turns, voice, scheduled prompts, and kanban runs. Switching CLI clears the model pin. Changing the Settings default does not rewrite chats that already have a harness.
| CLI | How a turn runs |
|---|---|
| Claude Code | Default. Streamed print mode. Sign in with claude /login, or set a key under Connections. |
| Codex | codex exec in the workspace. A later turn resumes that thread. Sign in with codex login. |
| Cursor | cursor-agent in print mode, with workspace trust so the TUI does not block a piped run. |
| Gemini | Gemini CLI over the Agent Client Protocol. Brompter refuses a CLI older than 0.33.0. |
| Orca | Optional. Settings → Agent can launch the turn through Orca in that workspace. |
Models
The model list comes from that CLI’s catalogue. Refresh asks the CLI again. A blank model follows Settings for the chat’s CLI. Auto-naming uses the same CLI that ran the first turn, with a cheap default (Haiku, a cheap Codex id, or Composer) unless you override it.
Portraits and status
Each chat has a small head. The helmet follows the harness, and a stable detail from the chat keeps two chats from sharing a face. The same head sits on replies, the panel header, and Recent.