CloudNavi
← Back to articles
Paseo Review 2026: Orchestrate Claude Code, Codex & OpenCode from Your Phone
AI Agents·2 min read
#Paseo#AI coding#orchestrator#Claude Code#Codex#OpenCode#mobile#Git worktree#AGPL#open source

Summary

That wish is exactly what **Paseo** delivers.

Paseo Review 2026: Orchestrate Claude Code, Codex & OpenCode from Your Phone


"I run Claude Code at my desk, but what if I could just tell it 'merge that PR' from my phone while commuting?"

That wish is exactly what Paseo delivers.

Paseo is an open-source orchestrator that lets you control 35+ AI coding agents — Claude Code, Codex, Copilot, OpenCode and more — from desktop, mobile (iOS/Android), web, or CLI, all from anywhere.

Paseo dashboard

Licensed under AGPL-3.0 and fully open source, with 11.1k stars on GitHub (as of July 2026).

In this guide we break down what Paseo actually does, how the mobile control works, how it compares to OpenCode / Claude Desktop / Codex App, and who should (and shouldn't) use it.

By the end you'll know:

  • The full picture of what Paseo can do
  • How agents run from your phone
  • Parallel multi-agent execution (Git worktree isolation)
  • How it differs from OpenCode / Claude Desktop
  • Setup steps and who it's best for

What is Paseo, exactly?

Paseo is a daemon (always-on process) + client suite for operating AI coding agents "from anywhere, all at once."

In one line: "A window into your PC/server's Claude Code or Codex, remotely controllable from your phone or another PC."

Architecture (how it works)

Paseo runs a daemon on your machine (PC, VM, dev server). Into it connect:

  • Desktop app (macOS / Linux / Windows)
  • Mobile app (iOS / Android)
  • Web app
  • CLI

all over WebSocket. The daemon launches and manages your installed claude codex opencode CLIs as local processes, using your own credentials.

Meaning: "The agent itself runs in your environment. Paseo just makes it reachable from anywhere."


What Paseo can do (feature list)

Control agents from your phone

  • Native iOS / Android apps
  • Start a task at your desk, then from the train ask "did that branch's tests pass?"
  • Same UI and workflow as desktop

35+ agents supported

  • Native: Claude Code, Codex, OpenCode, Pi
  • Via ACP catalog (+34): GitHub Copilot, Cursor, Gemini CLI, Amp, and more
  • Custom: add any CLI agent
  • Agent Client Protocol (ACP) compliant — any ACP agent connects

Parallel multi-agent execution (worktree isolation)

  • Each agent isolated in its own Git worktree
  • Run in parallel with no port conflicts
  • Example: web.fix-auth.my-app.localhost web.add-search.my-app.localhost — each worktree gets its own dev server URL automatically

Split panels

  • Agent / Browser / Terminal / Diff side by side
  • ⌘D (split vertical) · ⌘D (split horizontal) · ⌘W (close) · ⌘N (new agent)
  • ⌘K command palette

Freedom from port management

  • Each worktree's dev server gets a unique localhost subdomain
  • "Port 3000 already in use" problem disappears

Review, preview, ship (GitHub integration)

All in-app:

  • Commit / push
  • Create PR / watch checks / review
  • Merge
  • Inline diff review (changed files, line counts)

⌨ Keyboard-first

  • Panel switch ⌘1-9
  • Command palette ⌘K
  • Full keyboard operation

Voice control (fully local)

  • Dictation and real-time voice mode
  • Speech-to-text (STT) and text-to-speech (TTS) processed on-device (privacy-preserving)

Fully scriptable (CLI / automation)

# Delegate OAuth implementation to OpenCode
paseo run --provider opencode "implement OAuth"

# Refactor in a worktree with Claude Code
paseo run --provider claude --worktree refactor-auth "refactor auth"

# Connect to a remote daemon and run tests
paseo run --host devbox:6767 "run the test suite"

# List running agents
paseo ls

# Send follow-up instruction to a specific agent
paseo send <agent-id> "add tests"

# Auto-run codebase audit every Monday 9am
paseo schedule create --cron "0 9 * * 1" "audit the codebase"
  • paseo loop: auto-retry the agent until the verification command passes
  • MCP server: other agents can create worktrees, launch agents, open terminals, send prompts

Privacy-focused

  • No telemetry, no tracking, no forced login
  • Daemon runs on infrastructure you control
  • Your code and context never leave your machine

Comparison with similar tools

Here's where Paseo sits against the tools it's most compared to.

| Feature | Paseo | OpenCode Desktop | Claude Desktop | Codex App | |---------|-------|------------------|----------------|-----------| | License | AGPL-3.0 (OSS) | OSS | Proprietary | Proprietary | | Desktop | mac/Win/Linux | mac/Win/Linux | mac/Win | mac/Win | | Native mobile | iOS/Android | None | None | None | | Agents supported | Claude/Codex/OpenCode/Pi +35 | OpenCode only | Claude Code only | Codex only | | Git worktree isolation | Standard | None (sessions only) | None | None | | Per-worktree dev URL | Yes | No | No | No | | Split panels | Agent/Browser/Terminal/Diff | Sessions | Chat-centric | Terminal | | In-app browser | Yes | No | No | No | | GitHub workflow | Commit/PR/Checks/Review/Merge | GitHub integration | Integration | Integration | | CLI | run/ls/send/schedule/loop | OpenCode CLI | Limited | Limited | | MCP server | Yes | Bundled | Yes | No | | Local voice | Yes | No | No | No | | Self-hosted daemon | Yes | Server/local | No | No |

The biggest difference is "multi-agent orchestration" and "mobile." OpenCode / Claude Desktop / Codex App are apps for "one agent" — Paseo manages "all of them on one screen."


Who should use Paseo (and who shouldn't)

[Yes] Best for

  1. People juggling multiple agents

    • Implement with Claude Code, review with Codex, fix with OpenCode — in parallel
  2. People who want to check/control coding from their phone

    • Review a PR's checks on your commute, say "add tests" while waiting
  3. People using a remote dev server (VPS / Devbox)

    • No GPU needed on your local PC. Run agents on the server, control from your phone
  4. Privacy-conscious users

    • Your code is never sent to an external cloud. The daemon is your own infrastructure
  5. People who want automation / scripting

    • paseo schedule for weekly audits, paseo loop for auto-retry until tests pass

[No] Look elsewhere if

  1. You only use one agent

    • If Claude Code alone is enough, Paseo's overhead isn't worth it
  2. You don't need mobile, desktop-only

    • OpenCode Desktop is sufficient
  3. You prefer a proprietary all-in-one IDE

    • If Cursor-style is your taste, use that
  4. You're uncomfortable with CLI agents

    • Some technical knowledge (SSH, Git worktree, daemon management) is required

Setup steps (quick overview)

Step 1: Install the daemon

# Global install via npm
npm install -g paseo

# Or use Docker / desktop app (Linux/macOS/Windows)

Step 2: Prepare agent CLIs

  • Pre-install and authenticate claude codex opencode etc.

Step 3: Start the daemon

paseo daemon

Step 4: Connect a client

  • Connect from desktop app, web app, or mobile app
  • Auto-detected on the same machine; for remote, specify with --host

Time to finish: ~10 minutes if you already have the CLI and agents installed.


FAQ

Q1: Is it free?

Free and open source (AGPL-3.0). No cost for Paseo itself. API costs for each agent (Claude Code etc.) are separate.

Q2: Does the mobile app really work?

Yes. Native apps on both iOS and Android, with the same workflow as desktop.

Q3: Which agents can I use?

Claude Code, Codex, OpenCode, Pi are native. Via ACP catalog: Copilot, Cursor, Gemini CLI, Amp and +34 more. Custom CLI also addable.

Q4: Is my code sent externally?

No. The daemon runs on your machine, agents are local processes. No telemetry either.

Q5: Can I run multiple agents at once?

Yes. Each agent isolated in a Git worktree, no port conflicts in parallel execution.

Q6: Can I put the daemon on a VPS?

Yes. "Run agents on a remote machine" is the base configuration. You connect from your phone or another PC.

Q7: How is it different from OpenCode?

OpenCode is an app for "one agent (OpenCode)". Paseo manages "Claude Code, Codex, OpenCode — all of them" on one screen, with mobile, worktree isolation, and script automation.

Q8: Does voice input support English?

Yes — it depends on the local STT engine. Since processing is on-device, an English STT model gives you English voice control.


Conclusion — Should you use Paseo?

Verdict: if you want to run multiple agents "from anywhere," it's the strongest option of 2026.

  • Manage 35+ agents on one screen
  • Native control from phone (iOS/Android)
  • Parallel development via Git worktree isolation, no port conflicts
  • Fully open source (AGPL-3.0), privacy-first
  • Advanced automation via CLI / schedule / loop / MCP

On the other hand, wait-and-see if:

  • You only use one agent
  • You're not interested in mobile control
  • You're uncomfortable with server management / CLI

Paseo is the "cockpit for AI coding." Even away from your desk, your agent squad waits at your fingertips.

"Orchestrate coding agents from your desk and your phone."

From your desk, from your phone — your agents, from anywhere. That's Paseo.


Information in this article is as of July 2026. Features and specs may change without notice.