FR
Reference

Command reference

Reference for commands and options available in the Palabre CLI.

This page lists the main Palabre commands.

Startup commands

CommandDescription
palabreOpens the TUI home screen, creates the global configuration on first launch, and refreshes detected known agents.
palabre initExplicitly creates the global configuration and detects agents.
palabre init --localCreates a configuration in the current folder.
palabre doctorDisplays a diagnostic by sections: configuration, local tools, agents, and items to check.
palabre agentsLists declared agents and their local detection.
palabre -aShortcut for palabre agents.
palabre history, palabre historiqueLists recent Palabre Markdown exports.
palabre context scan --jsonPreviews the project context Palabre would retain.

Configuration

The TUI home screen performs the same conservative known-agent synchronization as palabre config --sync-agents: it adds newly detected known agents and refreshes known command names while preserving custom agents and user defaults.

CommandDescription
palabre configOpens the configuration assistant.
palabre config --set-defaults codex claudeSets the default agents.
palabre agent-role claude criticSets one agent persistent role.
palabre config -t 4Sets the default number of turns.
palabre config --summary-agent claudeSets the default summary agent.
palabre config --summary-agent noneRemoves the default summary agent.
palabre config --ask-summary-agent opencodeSets the default summary agent for ask mode.
palabre config --mode chatSets the default mode (chat, debate, or ask).
palabre config --ask-agents codex claude opencodeSets the default ask agents, 4 maximum.
palabre config --interface tuiSets the default interface (tui or terminal).
palabre config --language enSets the Palabre and agent prompt language.
palabre config --clear-defaultsRemoves default settings.
palabre config --sync-agentsAdds missing detected agents and refreshes known commands.
palabre config --ollama-models --jsonLists installed Ollama models and the configured model status.
palabre config --set-ollama-model <model>Sets the model for the ollama-local agent.
palabre config --sync-ollama-modelReplaces the configured Ollama model if it is no longer installed.

Debates

CommandDescription
palabre newOpens the guided assistant for a debate or Ask request.
palabre -s "Subject" -t 4Launches with default agents.
palabre codex-claude "Subject" -t 4Launches with a preset.
palabre run --subject "Subject" --agent-a codex --agent-b claudeLaunches with explicit agents.
palabre run --subject "Subject" --agent-a codex --agent-b claude --role-a architect --role-b criticLaunches with temporary roles.
palabre ask "Subject" --agents codex claudeRuns a request with independent responses.
palabre chat --agent-a codexOpens a continued conversation with one agent.

Conversation with one agent

palabre chat --agent-a <agent> opens a terminal conversation. The first message becomes its initial context; "Subject" remains accepted as an optional way to prefill that context. For every message, Palabre starts a new call to the selected CLI and injects the accumulated history into it. The result can therefore remain coherent in the current session without relying on a persistent interactive session inside Codex, Claude, or another tool.

Unlike Ask, which collects one or more independent answers to a single question before an optional summary, Chat maintains a multi-turn conversation with one active agent.

Use /agents to show available agents. /end saves the conversation in .palabre/, shows its file and folder, then returns home; /home returns without saving. /consult <agent> explicitly requests a second opinion on the six most recent retained messages; that opinion is added to the conversation. /use <agent> then switches the active agent, or you can continue with the initial agent. --role-a, --model-a, --language, --files, and --context remain available. Without --agent-a, Palabre uses the configured default agent A.

This first version remains stateless and performs no action. The context sent with each call is bounded to the six most recent messages; Palabre reports when older messages are no longer sent. Consultation remains explicit: an agent may suggest it, but the user keeps the initiative.

After a Debate or Ask completes in the TUI, /chat directly opens a conversation whose context shows its provenance and carries over the subject with the final summary, or the six recent exchanges when no summary is available.

TUI Commands

These commands are available from the TUI home screen or from /config depending on context.

CommandDescription
/chatOpens a conversation with the default agent A. /home or /exit returns to the home screen.
/askSwitches the TUI home screen to Ask mode.
/debatSwitches the TUI home screen to debate mode.
/agentsDisplays available agents or updates active agents when names are provided.
/rolesDisplays available roles or updates active roles when roles are provided. In Ask mode, a single role such as /roles critic applies to all Ask agents.
/configOpens TUI settings.
/historyShows recent Markdown exports. Alias: /historique.
/updateChecks the npm version and displays installation-specific update instructions.
/ollamaDisplays the configured Ollama model and installed models from /config.
`/ollama-url <urldefault>`
/ollama-model <model>Changes the configured ollama-local model from /config.
/ollama-syncReplaces the configured Ollama model with an installed model when the current one is missing.
/language fr, /language enChanges the language from /config. Alias: /lang.
/interface tui, /interface terminalChanges the default interface from /config.
/homeReturns to the TUI home screen. Alias: /back.
/quitExits the TUI.

Integrations

An integration launching Palabre can declare its identity in the child process environment:

PALABRE_CLIENT=palabre-vscode
PALABRE_CLIENT_VERSION=1.7.1

Palabre records this information, together with its own version, in .debate.md, .chat.md, and .ask.md exports. Client names are open so future integrations can identify themselves without a hard-coded registry. When no client is declared, the invocation source is direct-cli.

CommandDescription
palabre agents --jsonLists configured agents, their role, type, and availability as JSON v1.
palabre presets --jsonLists presets and their local availability.
palabre history --jsonLists recent Markdown exports as JSON v1.
palabre config --ollama-models --jsonReturns local Ollama state as JSON v1 for integrations.
palabre context scan [paths...] --jsonReturns the folders, files, and warnings from the --context scan as JSON v1.

General options

OptionDescription
-h, --helpDisplays help.
-v, --versionDisplays the version.
--config <path>Uses an explicit configuration file.
--trust-configTrusts the local configuration fingerprint after its contents have been reviewed.
--language <fr|en>, --lang <fr|en>Forces the Palabre and agent prompt language for this command.
--tuiForces the TUI interface, even if configuration asks for terminal rendering.
--terminal, --no-tuiUses raw terminal rendering. Also useful with palabre doctor for logs.
--plainHistorical alias for --terminal.
--renderer <auto|pretty|plain|tui|ndjson>Selects the terminal or integration renderer. tui enables the full-terminal interface.

Debate options

OptionDescription
-s, --subject <text>Session subject.
--topic <text>Compatible alias for --subject.
--preset <name>Selects an agent pair.
--agent-a <name>First agent.
--agent-b <name>Second agent.
--role-a <role>Temporary role for the first agent, without editing config.
--role-b <role>Temporary role for the second agent, without editing config.
--ask-role <role>Temporary shared role applied to all Ask agents.
--mode <debate|ask>Selects the orchestration mode.
--agents <names...>Ask mode agents, 4 maximum.
-t, --turns <number>Total number of turns, between 1 and 20.
--no-early-stopDisables early stopping.
--model-a <model>Model passed to agent A.
--model-b <model>Model passed to agent B.
--ollama-url <url>Overrides the address of every Ollama agent for this session.
--pull-modelsAllows Ollama to download a missing model.
--summary-agent <name>Summary agent.
--summary-model <model>Summary model.
--no-summaryDisables the final summary.
--files <paths...>Injects specific files.
--context <paths...>Scans text files or folders.
--show-promptDisplays the first turn prompt without calling any agent.
--dry-runPreviews the resolved session without calling agents or writing an export.

Update

CommandDescription
palabre updateDisplays update instructions.
palabre update --applyApplies the update from a git checkout.