Choose a mode
Palabre offers four modes: Debate, Relay, Ask and Chat.
| Mode | Agents | Exchanges | Summary | Export |
|---|---|---|---|---|
| Debate | 2 | Agents answer each other. | Yes, by default | .debate.md |
| Relay | 1 target conversation | One message to an existing conversation | No | .relay.md |
| Ask | 1 to 4 | Agents answer independently. | Yes: per-agent summary then comparison | .ask.md |
| Chat | 1 active, optional consultation | The user talks with the active agent. | No | .chat.md |
Debate first
Choose Debate to challenge a decision, find blind spots, or obtain a recommendation built from explicit disagreement.
/debate
/agents codex claude
Review this architecture and recommend a decision
See the Debate mode guide.
Relay to hand off
An agent sends a message to an existing Codex or Claude Code conversation and receives its reply. This direct command has no TUI screen; its Mission option keeps the scope and budget for several read-only steps. See Relay.
Ask to compare without influence
Choose Ask to collect up to four independent answers before comparing them. See the Ask guide.
For human use, start with palabre and use the TUI. Extensions should consume the integration contracts, not parse TUI output.
Chat to continue
Choose Chat when one agent is enough, with a second opinion through /consult. After Debate or Ask, it can carry over the subject and summary. See the Chat guide.
Resume without replaying answers
Add --checkpoint to a Debate or Ask request when a session should survive an interruption. Palabre records state under .palabre/sessions/; palabre sessions lists it and palabre resume <session-id> starts only the next step. Resume validates configuration and context before calling an agent; see the CLI reference for confirmations and targeted deletion.
See Summaries for the Debate and Ask formats.