The Problem

Developers who use multiple coding agents (Claude Code, Codex, OpenCode) end up managing several terminal sessions across different machines. There's no unified way to start, monitor, or interact with agents from a phone or a different computer. Each agent has its own CLI, its own session management, and no remote access story.

What This Does

Paseo runs a local daemon that manages coding agents on your machine. Clients—desktop, mobile, web, and CLI—connect to it over the network. The packages/app directory contains a React Native/Expo client that runs on iOS, Android, and web. The daemon itself handles agent lifecycle, session state, and remote connections.

The CLI (@getpaseo/cli) provides terminal access: paseo run starts agents, paseo ls lists them, paseo attach streams live output, and paseo send sends follow-up tasks. Remote daemon support means you can run paseo --host workstation.local:6767 to control agents on another machine. The packages/app/src/components/ directory holds the UI components for agent status, workspace management, and file exploration.

How To Use It

Setup: Install the CLI globally and start the daemon: npm install -g @getpaseo/cli paseo

This prints a QR code. Scan it from the mobile app to pair your phone.

Configuration: You need at least one agent CLI installed and authenticated—Claude Code, Codex, or OpenCode. The daemon discovers them from your environment. No API keys or environment variables are required beyond what your agent CLIs already use.

Running it: From the terminal, you can run agents directly: paseo run --provider claude/opus-4.6 "implement user authentication" paseo ls paseo attach abc123 paseo send abc123 "also add tests"

The desktop app (from paseo.sh/download) starts the daemon automatically—no manual setup needed.

Real-World Use

You're working on a feature at your desk with Claude Code. You need to leave for a meeting but want to check progress from your phone. Open the Paseo mobile app, see the agent's current status, and send a follow-up: "add error handling for the auth endpoint." When you return, the agent has completed the task and committed the work. For parallel work, you can start a Codex agent on a separate worktree (--worktree feature-x) while Claude handles the main branch.

Code Health & Issues

The repo is well-structured overall. Tests exist across the codebase (30 test files, including E2E Playwright specs in packages/app/e2e/), CI is configured in .github/workflows/ci.yml, and there's a license. The codebase is TypeScript throughout with clear separation between app, CLI, and daemon concerns. Med - The packages/app is a monorepo containing both React Native and web builds. The react-native-web.d.ts and metro.config.cjs suggest cross-platform complexity that can introduce platform-specific bugs. The E2E test suite covers many paths, but mobile-specific flows (Maestro tests) are less comprehensive. Low - The CLAUDE.md and AGENTS.md files indicate the repo is designed to be worked on by AI agents. This is fine, but it means some documentation may assume AI-agent context that human contributors lack. Low - The orchestration skills (npx skills add getpaseo/paseo) are explicitly marked as unstable and "use at your own risk." The README notes they change frequently and may be coupled to the author's setup.

The Bottom Line

Paseo solves a real problem: unified control of coding agents across devices. The architecture is sound—a local daemon with multiple client types is a pragmatic design. The CLI is genuinely useful for headless/remote workflows. It's early-stage software with a single primary author (13.5k stars on the upstream repo suggests real traction), so expect rough edges in the experimental features. Best suited for developers already using Claude Code or Codex who want phone-based oversight and multi-machine orchestration.