FR
Get Started

Introduction

Understand what PALABRE does, how it orchestrates your AI agents, and what privacy guarantees it provides.

PALABRE lets the AI agents already installed on your machine cooperate to make better decisions before acting. You keep your tools, subscriptions and models; PALABRE organizes the exchanges and keeps a Markdown record.

Four modes, depending on your need

  • Debate: two agents respond to each other, then a summary sets out agreements and disagreements.
  • Relay: an agent sends a message to an existing Codex or Claude Code conversation and receives its reply, without copy-paste from you.
  • Ask: up to four agents answer independently, then a summary covers each answer and compares them.
  • Chat: you talk to one active agent and can request a second opinion with /consult.

For several read-only steps within an agreed scope and budget, Relay offers the Mission option. You make decisions; the mission lead calls the teammate one explicit step at a time.

What is PALABRE for?

Compare approaches, challenge a plan, confront opinions, or hand off to a conversation that already knows your project. Debate and Ask can produce a final summary, disabled with --no-summary; Chat and Relay do not add one.

What PALABRE does

  • detects installed agents and opens a TUI with palabre in an interactive terminal;
  • supplies the subject, context and useful history to Debate, Ask and Chat calls;
  • can send selected files with --files or --context;
  • makes Debate or Ask resumable with --checkpoint and palabre resume;
  • relays a message to an existing conversation with palabre relay;
  • exports exchanges as .debate.md, .relay.md, .ask.md or .chat.md, and the Mission option as .mission.md.

What stays under your control

PALABRE does not create accounts or provide model access. Each agent must be installed and, where required, authenticated separately. A role or a message from another agent never authorizes an action on your behalf.

Debate, Ask and Chat make a new agent call for each reply. PALABRE supplies the whole Debate transcript, or at most the six recent Chat messages. Relay instead resumes an existing conversation and can modify its history: closed by default, or open with --open for Codex on Windows, macOS, or Linux (pilot; python3 required on POSIX). It never resends a message automatically.

Privacy

The PALABRE CLI does not collect your data: no PALABRE account, no CLI telemetry, no PALABRE server receiving your conversations. The CLI runs on your machine and saves its exports locally; these can contain sensitive information, so review them before sharing.

With a local Ollama server and local models on your machine, Debate, Ask and Chat can keep their AI exchanges local. All participants and any summary agent must use local models. Launch the session directly from your terminal if you do not want a remote host assistant to receive your request or the results.

If you include remote tools or models, they receive the necessary prompts, files and exchanges, subject to their policies and your settings. A remote Ollama server also receives the data sent to it. A cloud model used through Ollama is still remote, even when the Ollama server runs on your machine. Using just one local participant does not make the whole session local.

“Local AI exchanges” does not mean “no network requests”: PALABRE can, for example, query the npm registry to check for updates. This check does not send your prompts, files or conversations.

Next

Start by installing PALABRE, then run your first session. To hand off, read Relay; for settings, read configuration.