Introduction

Palabre is a CLI/TUI orchestrator for AI agents already installed on your machine.

The goal is simple: you give a subject, Palabre orchestrates the agents, displays their responses in the terminal, then exports a complete Markdown record. Debate and Ask modes can also produce a final summary.

Three modes are available:

• chat: you work with one active agent and can explicitly request a second opinion with /consult;
• debate: two agents take turns discussing a subject;
• ask: several agents answer the same request independently, then a summary agent faithfully summarizes each response and compares them.

Palabre does not replace Claude Code, Codex CLI, Antigravity CLI, OpenCode, Mistral Vibe, or Ollama. It drives them. You therefore keep your tools, subscriptions, default models, and terminal habits.

Palabre runs locally on your machine. It does not send any data to a server owned by Palabre. Prompts, files, and transcripts are only transmitted to the agents you choose to use.

Privacy therefore depends on the agents selected. Before sending code, documents, or sensitive data to Claude Code, Codex CLI, Antigravity CLI, OpenCode, Mistral Vibe, Ollama, or any other configured agent, check their own privacy policies and your account settings.

What is Palabre for?

Palabre is useful for:

• maintaining a continued conversation with one agent and requesting a second opinion without losing context;
• comparing two agents on the same question;
• comparing several independent answers with Ask mode;
• getting a critical review of an idea;
• pitting a local Ollama model against a more powerful CLI;
• producing a shareable Markdown transcript;
• turning an AI discussion into consensus, disagreements, proposed actions, and conclusion.

What Palabre does

Palabre:

• detects CLIs available on your machine;
• opens a TUI by default with palabre in an interactive terminal;
• launches agents in batch mode;
• injects the subject, session context, and useful history into each call;
• maintains a stateless Chat with one active agent and bounded history;
• can run Ask requests without conversation between agents;
• can add project files or folders to the context;
• generates a final summary;
• exports the session to a .debate.md, .chat.md, or .ask.md file.

What Palabre does not do

Palabre does not provide paid access to models. It does not create a Claude, OpenAI, Google, or Ollama account. Each agent must be installed and authenticated separately.

Palabre also does not keep an interactive session open with the CLIs. Each response is produced by a batch call. Continuity comes from the transcript Palabre injects: the complete debate history in Debate mode, and at most the six most recent messages in Chat.

Next

Start by installing Palabre, then run your first session. The configuration guide explains the details when you want more control.