Paseo: AI Coding Agent Orchestration
/paseo
Paseo AI coding agent is a thin wrapper that unifies Claude Code, Codex, Copilot, OpenCode, Pi, Antigravity, and Muse Code through one daemon. You get parallel worktrees, cross-device access, and agents that can spawn other agents — but that orchestration layer adds latency on every handoff, and when two agents overwrite the same file you get to manually sort it out. The real sweet spot is teams already running multiple providers who want one UI instead of five terminals.
What is Paseo?
Paseo is a self-hosted agentic development environment for desktop, mobile, web, and CLI. It runs a local daemon that coordinates multiple AI coding agents — Claude Code, Codex, Copilot, OpenCode, Pi, Antigravity, and Muse Code — in parallel, each in its own worktree. You get one interface instead of one per provider, with built-in orchestration so agents can delegate work to each other.
Key features
- ✓Run multiple AI coding agents simultaneously, each in its own isolated worktree on one machine or distributed across machines — all through one unified interface.
- ✓Pick the right model per job from Claude Code, Codex, Copilot, OpenCode, Pi, Antigravity, and Muse Code without switching tools.
- ✓Self-hosted: agents run on your machine with your dev environment, your configs, your credentials — no forced cloud dependency.
- ✓Zero telemetry, zero tracking, zero forced logins. The daemon is yours; your code never leaves your network unless you explicitly relay it.
- ✓Cross-device: start a session on desktop, check progress from iOS or Android, script it from the CLI — all synced via end-to-end encrypted relay or direct TCP.
- ✓Agents can create worktrees, launch other agents, and hand off tasks — you define the committee, Paseo handles the logistics.
- ✓Full dev workflow in split panes: file editor, terminals, diff viewer, pull requests, and a browser — no tab-switching.
- ✓Plugin system written in TypeScript with access to UI components, agent lifecycle hooks, and provider integrations — write once, runs on every client.
Installing Paseo: desktop app, CLI, Docker, and mobile
The desktop app is the fastest path on macOS, Windows, or Linux — download from the releases page and the daemon starts automatically. For a server or headless setup: `npm install -g @getpaseo/cli` then `paseo`. Docker is one command: `docker run -d --name paseo -p 6767:6767 -e PASEO_PASSWORD=change-me -v "$PWD/paseo-home:/home/paseo" -v "$PWD:/workspace" ghcr.io/getpaseo/paseo:latest`. iOS and Android apps are full Paseo clients — pair them to your daemon via Settings, over an end-to-end encrypted relay or direct TCP.
What you can build with Paseo
- •Parallel code review: spin up separate agents across worktrees to review PRs, run migrations, or write tests concurrently, then review diffs side-by-side in the same window.
- •Orchestrated handoff: have a planning agent delegate subtasks — research to one agent, implementation to another, security review to a third — and pull results back together.
- •Multi-provider fallback: use Claude for reasoning-heavy work and Codex for speed-sensitive tasks on the same codebase, switching providers mid-conversation.
- •Remote workstation: run the daemon on a headless server or VPS, connect your mobile app over an end-to-end encrypted relay, and code from anywhere without tunneling your whole environment.
- •Plugin-powered dashboards: build issue integrations or orchestration services with the TypeScript SDK, running as a long-lived process connected to your daemon over WebSocket.
Strengths
- ✓Single interface for seven providers means one onboarding path instead of seven, and one way to monitor running agents.
- ✓Agents run locally on your hardware with your tools — no vendor lock-in and no sending context to a third-party relay unless you opt in.
- ✓Daemon architecture means one process to manage but one hub that every client — desktop, mobile, web, CLI — connects to for consistent state.
- ✓Cross-device relay is end-to-end encrypted and optional, so mobile access doesn't require exposing your network or using a third-party tunnel service.
- ✓Plugin system exposes real APIs — UI panels, lifecycle hooks, provider integrations — rather than just config flags.
- ✓Built by one person with an active community: the scope is coherent and the UX is opinionated rather than committee-designed.
Paseo limitations and considerations
- △Multi-agent orchestration adds complexity: agents can overwrite each other's files, and handoff between providers introduces latency that a single-agent workflow avoids entirely.
- △The daemon is another process to monitor, log, and maintain — on a team without dedicated DevOps this is operational overhead you trade for convenience.
- △Plugin API is still maturing; real-world plugins are sparse and you may hit gaps or breaking changes as the system evolves.
- △Each provider requires its own CLI installed and configured — the multi-provider story is real but setup is not one click.
- △One-person maintenance is impressive but carries risk: long-term roadmap, bug response time, and feature velocity depend entirely on one person sustaining the project.
Paseo alternatives
Frequently asked questions about Paseo
Yes. Paseo is Apache-2.0 licensed, so it is free for personal and commercial use. The README also mentions GitHub Sponsors as a funding path for the independent maintainer.
Yes. Paseo has native apps for both iOS and Android that connect to your daemon over an end-to-end encrypted relay or directly over TCP when on the same network.
Paseo supports Claude Code, Codex, GitHub Copilot, OpenCode, Pi, Antigravity, and Muse Code — seven providers accessible through one interface.
Paseo collects no telemetry, runs no tracking, and requires no forced account. The daemon is entirely local, and device pairing uses end-to-end encryption.
Yes. Install the CLI with `npm install -g @getpaseo/cli` and run `paseo`, or use the Docker image for a one-container setup. Connect from any client over WebSocket.
Plugins are written in TypeScript, run on the daemon, and appear in every connected client. They have access to UI components (panels, sidebar items, themes), agent lifecycle hooks, and the ability to add new providers. Install from the registry, Git, or a local directory.
The problem it solves
Separate agent CLIs mean separate windows, manual context handoffs, and no unified view of what each agent is doing. You can't easily run Claude Code and Codex on the same task, hand off mid-conversation, or review four agents' diffs side-by-side without a custom tmux layout. A solo developer doing a sprint might juggle three terminal tabs plus a browser, context-switching constantly just to coordinate agents that ought to know about each other.
Who should try it — and who should skip
Paseo makes sense if you're already running or evaluating multiple AI coding agents and want one interface instead of a collection of terminal windows. It also helps if you need parallel agents working simultaneously on a codebase or want cross-device access without rolling your own daemon. Solo developers with a single provider and straightforward tasks can skip it — the daemon overhead only pays off once you actually need orchestration. Teams should weigh whether the single-maintainer maintenance model fits their risk tolerance before betting on it for critical paths.
Related repositories
Is paseo worth your time?
ChatGPT, Claude and Perplexity can all read this page. Ask one of them what it makes of paseo.