Chat mode
Chat fits situations where one agent is enough, with the option to consult another agent when useful.
From the TUI
Run palabre, then /chat. Chat uses default agent A. /agents codex selects the active agent.
| Command | Effect |
|---|---|
/consult claude | Requests a second opinion without replacing the active agent. |
/use claude | Continues with Claude. |
/agents | Displays available agents. |
/end | Saves and ends. |
/home | Returns without saving. |
After Debate or Ask, /chat carries over the subject and final summary, or the six recent exchanges without a summary.
Memory and limits
Chat is stateless for external agents. Every response starts a batch call and reinjects at most six recent messages. It does not promise a persistent provider session, token streaming, or recovery after restart.
Direct command
palabre chat --agent-a codex
--role-a, --model-a, --language, --files, and --context remain available. The .chat.md export preserves the transcript, time, and termination reason, without an automatic summary.
In the TUI, with the direct command, and in an NDJSON integration, a blank line does not end the conversation. /end saves it; /exit, /home, or the end of input finish it without an export. If an agent fails, Palabre saves the partial transcript. --dry-run is not available in Chat: the command is refused without running anything.
Integrations drive Chat through the NDJSON stream.