Debate and Ask

TUI home screen

palabre

The TUI home screen is the recommended starting point. It shows active agents, roles, summary, current folder, and useful commands.

From this screen:

• type a subject to launch the current mode;
• use /ask or /debat to switch mode;
• use /agents to choose agents;
• use /roles to apply roles;
• use /history to find recent exports;
• use /config to change settings without leaving the TUI.

palabre new remains available as the historical guided assistant.

Debate mode

debate mode makes two agents talk turn by turn. It is the right choice when you want to confront two viewpoints, build real contradiction, request a critical review, or turn a fuzzy topic into a decision.

Each turn receives:

• the subject;
• the session context;
• files or folders injected with --files or --context;
• the history already produced by the agents.

At the end, Palabre asks for a final summary. This summary covers the consensus, disagreements or uncertainties, proposed actions, and a short conclusion.

With default agents

palabre -s "Critique this plan" -t 4

This command uses the agents defined in your configuration.

With a preset

palabre codex-claude "Critique this plan" -t 4

A preset quickly selects an agent pair. Order matters: codex-claude has Codex respond first, then Claude.

With explicit agents

palabre run --subject "Critique this plan" --agent-a codex --agent-b claude --turns 4

This form is longer, but makes the command fully explicit.

Ask mode

palabre ask "Compare these two options" --agents codex claude opencode

Ask mode sends the same subject and context to several agents, without a shared transcript between them. Each agent answers alone. Then the summary agent faithfully summarizes what each agent said and compares the responses.

Ask is useful when you want to compare several independent answers, get multiple formulations, check whether agents converge without influencing each other, or ask several profiles to analyze the same request in parallel.

Without --agents, Palabre uses defaults.askAgents if defined, otherwise the default debate pair.

Ask mode accepts 1 to 4 agents:

palabre ask "Which approach should we choose?" --agents codex claude opencode antigravity

The export uses the .ask.md extension.

--turns does not control the number of responses in Ask mode: the number of responses depends on the selected agents. To change the default Ask agents:

palabre config --ask-agents codex claude opencode
palabre config --ask-summary-agent opencode

From the TUI home screen, /ask switches to Ask mode, /agents chooses active agents, /roles applies roles, and /home returns to the home screen from secondary views.

Number of turns

--turns or -t sets the total number of turns, between 1 and 20.

Example with codex as agent A and claude as agent B:

By default, Palabre may stop before the limit if agents clearly express full agreement after a complete turn.

To force all requested turns:

palabre codex-claude "Subject" -t 4 --no-early-stop

Choose a model on the fly

Palabre does not list CLI models, as they change frequently. It simply passes the value to the agent. If the model does not exist or is not authorized by your account, the CLI will return the error.

palabre codex-claude "Subject" --model-a <codex-model> --model-b <claude-model>

For Ollama:

palabre codex-ollama "Subject" --model-b <installed-ollama-model>

Preview without calling AI

palabre codex-claude "Subject" --show-prompt

This command displays the first turn prompt, the agents, and the summary options, then stops. This is useful for verifying the context being sent without consuming an AI request.

TUI interface and raw terminal rendering

By default, Palabre favors the TUI interface when output is a real terminal. palabre opens the home screen, and a direct command such as palabre codex-claude "Subject" -t 4 runs the session with the TUI renderer.

To get raw rendering suitable for logs:

palabre codex-claude "Subject" --terminal

To make that the default behavior:

palabre config --interface terminal