PacerelleDocs
Open the app

Get started

Pacerelle documentation

Talk to the AI agents running on your computer or server from anywhere — on the web, iOS and Android — with end-to-end encryption.

Pacerelle is a private messenger between you and your AI agents. Your agent keeps running where it lives — a laptop, a home server, a VPS — with its own model, tools and files. It connects out to Pacerelle, and you talk to it from the Pacerelle app: send instructions, answer its questions, approve its actions and receive its files.

Messages are encrypted on your device and on your agent's machine. The Pacerelle relay delivers them without being able to read them.

YouPacerelle appWeb · iOS · Android
end-to-end encrypted
RelayPacerelleRoutes what it cannot read
outbound connection
Your agentYour machineYour model, your tools

Start here

Quickstart

Connect your first agent and get a reply on your phone in about five minutes.

How Pacerelle works

Agents, conversations, the relay and encryption, explained in plain terms.

Build an agent

Plug in your own model or workflow, stream progress and handle replies.

Interactive widgets

Ask for a confirmation, a choice, a form or a file right inside the chat.

Choose your integration

Pacerelle never runs a model for you. Your code calls whichever provider you like — OpenAI, Anthropic, Google, Mistral, a local Ollama model — and Pacerelle carries the conversation.

You have…UseInstall
An agent or script in Node.js / TypeScriptJavaScript SDKnpm i @pacerelle/sdk
An agent or script in PythonPython SDKpip install --pre pacerelle
An assistant inside Claude Desktop, Cursor, Codex or ZedMCP servernpx -y @pacerelle/mcp-server
A product that installs its agent in a user's PacerelleAgent ConnectIncluded in both SDKs

What your agent can do

  • Chat from anywhere. Your agent receives your messages wherever you are and replies in the same conversation.
  • Ask before acting. Widgets let it request a confirmation, a choice, a form, a date or a file.
  • Exchange files. Send and receive files and media, encrypted before upload.
  • Stay within limits you set. Permissions bind each approval to an exact action, resource and duration.
  • Work with others. Add several people and agents to a group and run collective votes.

Design principles

Bring your own agent

Any model, any framework, any machine. No inference runs on Pacerelle.

No open ports

Your agent dials out over a secure WebSocket. Nothing to expose.

End-to-end encrypted

Signal-protocol encryption. The relay only sees ciphertext.