Paseo — Summary

Paseo is a TypeScript monorepo that provides a unified management layer for multiple AI coding agent CLIs (Claude Code, Codex, Copilot, OpenCode, Pi) running on the user's own machines. A persistent local daemon orchestrates agent processes, exposes a WebSocket API, and relays connections from mobile (iOS/Android), desktop (Electron), web, and CLI clients. Users can run multiple agents in parallel on different worktrees, send tasks from their phone, and use voice dictation. Paseo's four skills (/paseo-handoff, /paseo-loop, /paseo-advisor, /paseo-committee) are Claude Code skills that teach agents to orchestrate other agents through the Paseo daemon — enabling multi-provider handoff, Ralph-style acceptance-criteria loops, advisory second opinions, and two-agent committee analysis. The project is self-hosted and privacy-first: no telemetry, no forced logins.

Differs from seeds: Closest to taskmaster-ai in providing a cross-tool management layer, but Paseo focuses on agent process orchestration (spinning up/monitoring concurrent CLI processes) rather than task breakdown. Unlike claude-conductor (markdown scaffold) or superpowers (skills pack), Paseo ships a full Expo mobile app, Electron desktop app, and CLI daemon. The skill system (/paseo-loop = Ralph loop, /paseo-committee = consensus advisory) borrows BMAD's committee pattern but implements it as runtime agent calls rather than persona markdown.

Paseo — Overview

Origin

Paseo is developed by Mouheb Boudra (@moboudra). The project's tagline: "Coding agents from your phone, desktop and CLI." It solves the problem of monitoring and controlling multiple AI coding agents across devices without giving up the dev environment setup you already have.

Philosophy

From README:

• Self-hosted: Agents run on your machine with your full dev environment. Use your tools, your configs, and your skills.
• Multi-provider: Claude Code, Codex, Copilot, OpenCode, and Pi through the same interface. Pick the right model for each job.
• Voice control: Dictate tasks or talk through problems in voice mode.
• Cross-device: iOS, Android, desktop, web, and CLI. Start work at your desk, check in from your phone, script it from the terminal.
• Privacy-first: No telemetry, no tracking, no forced log-ins.

Positioning

Paseo sits between the user and their existing AI agent CLIs. It does NOT replace Claude Code or Codex — it orchestrates them. The daemon manages the process lifecycle; clients (phone, desktop, web) are view layers.

Repo Facts

• GitHub: https://github.com/getpaseo/paseo
• Stars: 6,754 (2026-05-26)
• Language: TypeScript
• License: NOASSERTION (custom / proprietary terms likely)
• Last commit: 2026-05-26
• Community-built relay: https://github.com/zenghongtu/paseo-relay (Go)

Paseo — Architecture

Distribution

• Desktop app: Download from paseo.sh/download or GitHub releases (Electron)
• CLI/headless: npm install -g @getpaseo/cli && paseo
• Mobile: iOS/Android app via paseo.sh

Monorepo Packages

packages/
├── server/    # Paseo daemon (agent process orchestration, WebSocket API, MCP server)
├── app/       # Expo client (iOS, Android, web)
├── cli/       # `paseo` CLI for daemon and agent workflows
├── desktop/   # Electron desktop app
├── relay/     # Relay package for remote connectivity
└── website/   # Marketing site + documentation (paseo.sh)

Architecture Diagram

User Clients (phone / desktop / web / CLI)
       ↕  WebSocket / QR-code relay
    Paseo Daemon (packages/server)
       ├── Agent Process 1 (Claude Code --worktree feature-x)
       ├── Agent Process 2 (Codex gpt-5.4)
       └── Agent Process N (OpenCode ...)

Daemon Features

• Agent process orchestration (spawn, monitor, kill, resume)
• WebSocket API for all client types
• MCP server built into daemon (packages/server)
• QR code for mobile pairing
• Remote connectivity via relay

Supported Agent Providers

• Claude Code
• Codex (OpenAI)
• GitHub Copilot CLI
• OpenCode
• Pi

Required Runtime

• Node.js (for daemon + CLI)
• At least one agent CLI installed and configured
• npm for @getpaseo/cli

Config Files

• paseo.json — project-level Paseo configuration
• .mise.toml / .tool-versions — runtime version pinning
• CLAUDE.md / AGENTS.md — present in repo root for agent guidance

Paseo — Components

Skills (6, in skills/)

• paseo-handoff — Hand off work between agents (e.g., plan with Claude, implement with Codex)
• paseo-loop — Loop an agent against clear acceptance criteria (Ralph loop), optionally with a verifier
• paseo-advisor — Spin up a single agent as advisor for a second opinion without delegating work
• paseo-committee — Form a committee of two contrasting agents for root cause analysis + plan
• paseo-epic — Unknown (epic-level work breakdown)
• paseo-orchestrate — Orchestration coordination skill
• paseo — Core Paseo skill (base integration)

CLI (packages/cli)

Binary: paseo

Key subcommands:

• paseo run --provider <provider/model> "<task>" — Start an agent run
• paseo run --provider codex/gpt-5.4 --worktree feature-x "<task>" — Run with worktree
• paseo ls — List running agents
• paseo attach <id> — Stream live output
• paseo send <id> "<message>" — Follow-up to running agent
• paseo --host <remote>:<port> run "<task>" — Run on remote daemon

Desktop App (packages/desktop)

• Electron application
• Daemon starts automatically on open
• Shows running agents, output streams

Mobile App (packages/app)

• Expo (React Native) — iOS, Android, web
• Connect to daemon via QR code scan
• Voice control support

Daemon (packages/server)

• WebSocket API server
• Agent process lifecycle management
• MCP server integration
• Relay client for remote connections

Relay (packages/relay)

• Remote connectivity package
• Community Go implementation: paseo-relay by zenghongtu

Install Skill

npx skills add getpaseo/paseo

Installs all 7 skills into the project.

Paseo — Uniqueness & Positioning

differs_from_seeds

Paseo is closest to taskmaster-ai in providing a cross-tool management layer, but the comparison diverges quickly: taskmaster-ai manages a task decomposition tree with JSON state, while Paseo manages OS-level agent processes with real-time streaming. Paseo is also similar to claude-conductor in being multi-agent coordinator, but claude-conductor is zero primitives (markdown templates only), whereas Paseo ships a full daemon, Electron desktop, iOS/Android app, and CLI. The skill patterns (/paseo-loop = Ralph loop, /paseo-committee = two-agent advisory) borrow from the superpowers multi-agent pattern but implement them as daemon-driven subprocess orchestration rather than Claude Task tool calls.

Distinctive Positioning

1. Cross-provider skill orchestration: The only framework where skills explicitly cross provider boundaries — /paseo-handoff lets Claude plan and Codex implement within the same workflow. No other framework does this as a skill.

2. Mobile + voice as first-class: iOS/Android app with voice dictation for task input is unique in this corpus. The mental model is "manage your coding agents from anywhere."

3. Git worktree as isolation primitive: Explicit --worktree flag per run, multiple agents on separate worktrees, enabling true parallel execution without conflict.

4. Ralph loop as skill: /paseo-loop formalizes the acceptance-criteria loop pattern (popularized by smart-ralph/ralph-claude-code) as a reusable skill installable via npx skills add.

5. Remote daemon delegation: --host workstation.local:6767 enables treating a remote machine's agent environment as a compute target.

Observable Failure Modes

• License is NOASSERTION — unclear terms, may limit commercial use
• Requires all agent CLIs to be pre-installed and configured (prerequisite burden)
• Daemon is a dependency — if it crashes, all running agents are affected
• No container isolation — agents run natively; malicious code in project can affect host
• Mobile app requires relay for non-local access (additional infrastructure)
• Skill content not publicly readable in depth (skill source may be proprietary)