Paseo: Unified Interface for Managing Multiple Coding Agents Across Desktop and Mobile

What a coding agent is

A coding agent is an AI program that can operate a computer autonomously: reading and writing code, running tests, executing shell commands, and committing changes. Examples include Claude Code, Codex, and the GitHub Copilot CLI.

Each agent excels at a particular task (e.g., Claude Code for code modification, Codex for batch jobs), but the rapid proliferation of agents creates friction:

Developers must juggle multiple terminals or UI windows.

Task state is isolated per agent.

Coordinating several agents for a single workflow is cumbersome.

Running tasks are not visible on mobile devices.

Paseo: a unified orchestration layer

Paseo does not implement a new coding agent. It provides a management layer that takes already‑installed agents (Claude Code, Codex, Copilot, OpenCode, Pi, etc.) and offers a single interface for starting, stopping, assigning tasks, viewing status, and collecting results.

Architecture: daemon + multi‑client

The core consists of a persistent local daemon and one or more remote clients (desktop app, mobile app, web UI, CLI). The daemon schedules agents and manages their processes; clients act as remote controllers.

Create a task on any client, specifying the provider (e.g., Claude Code) and model.

The request is sent via WebSocket to the local daemon.

The daemon launches the corresponding agent CLI within the full local development environment.

Output from the running agent streams back to the client in real time.

Additional commands are sent with send, which the daemon forwards to the active agent.

The daemon also exposes a TypeScript SDK ( @getpaseo/client) for building custom orchestration services.

Key capabilities

Cross‑device control : Start a task on a desktop, then monitor or extend it from a phone, web page, or CLI (iOS, Android, desktop, web, CLI).

Parallel orchestration : Run multiple agents simultaneously, each handling a different sub‑task (e.g., one writes a backend API while another creates a frontend page); results can be merged later.

Voice input : The mobile client accepts spoken commands.

Privacy‑first execution : All agent processes run inside the local daemon; only encrypted control signals travel between devices, so code and credentials never leave the machine.

Getting started

Desktop app (recommended) : Download from the GitHub Releases page; the daemon starts automatically. Pair a mobile device by scanning the QR code in the settings.

CLI :

npm install -g @getpaseo/cli
paseo

The CLI prompts for end‑to‑end encrypted relays for device pairing; if declined, it falls back to TCP, Tailscale, or VPN.

Docker (self‑hosted web UI) :

docker run -d --name paseo -p 6767:6767 \
  -e PASEO_PASSWORD=change-me \
  ghcr.io/getpaseo/paseo:latest

Prerequisite: install and configure at least one agent CLI (Claude Code, Codex, Copilot, OpenCode, or Pi).

Typical CLI commands:

paseo run --provider claude/opus-4.6 "implement user authentication"

paseo ls                           # list running agents
paseo attach abc123                # view live output of an agent
paseo send abc123 "also add tests" # add a new instruction to a running agent

Comparison with single‑agent tools

Core positioning : single coding agent vs. multi‑agent orchestration layer.

Code generation : agents write code themselves; Paseo reuses existing agents.

Parallel support : not native in single agents; core capability in Paseo.

Cross‑device control : terminal‑centric for single agents; Paseo offers desktop, mobile, web, and CLI endpoints.

Code location : local machine for both; Paseo runs code inside the local daemon.

Dependencies : built‑in model for single agents; Paseo requires at least one external agent CLI.

Limitations

Paseo does not provide its own model capabilities; its effectiveness depends on the strength of the underlying agents.

Intended users

Developers already using multiple coding agents.

People who work on the move and need to monitor tasks from both desktop and mobile.

Users who need to run parallel tasks with different agents.

Teams that prioritize keeping code and credentials on‑premises.

Design ideas worth borrowing

Separation of scheduling and execution : a dedicated orchestration layer coordinates existing tools without duplicating functionality.

Local‑first + remote control : core computation stays on the developer’s machine; encrypted signals enable convenient remote interaction.

Provider‑agnostic interface : the same UI can schedule agents from any provider, allowing task‑driven model selection.

GitHub address: github.com/getpaseo/paseo