AI Commander vs Paseo
Paseo and AI Commander both let Claude Code and Codex run on hardware you own rather than in a hosted sandbox, but they solve different problems. Paseo is an app for running and steering many coding agents from your desk or your phone. AI Commander is the connection that lets the AI client you already use run commands and long jobs on your PCs, servers and GPU boxes. This page compares them honestly and shows where they fit together.
The short answer
What each one is
Paseo
In Paseo's words, it is "the agentic development environment": "Run many coding agents at once, on any machine.
From your desk and from your phone." A daemon on your machine manages the agents, and the desktop, mobile, web and
CLI clients connect to it. The desktop app bundles the daemon; on a server or VM you install
@getpaseo/cli and run it headless. Each agent can get its own git worktree, and the app includes an
editor, terminals, diffs, pull requests and a browser. Clients can connect directly over your local network; from outside it, Paseo offers an optional end-to-end
encrypted relay, your own tunnel such as Tailscale or Cloudflare Tunnel, or exposing the daemon port directly.
It is free and open source under Apache 2.0, and you bring your own agent CLIs and credentials.
AI Commander
A hosted relay plus a small agent per machine, reachable as an MCP server and as a plain REST API. It brings no
model and no chat window: the AI client you already use — Claude, ChatGPT, Codex, Grok or any MCP client — is
the interface, and AI Commander is the hands on your machines. The agent holds one outbound WebSocket open, so
there is no inbound port, no exposed SSH and no VPN, and it runs on macOS, Windows and Linux. Detached jobs
(remote_job_start, _status, _logs, _cancel) outlive the call
and the conversation. One of the things you can start that way is Claude Code or Codex itself — see
remote coding agents.
Side by side
| AI Commander | Paseo | |
|---|---|---|
| What it is | MCP server and REST API that your AI client uses to act on your machines | Desktop, mobile, web and CLI app to run and steer coding agents |
| Where you type | In the AI client you already use (Claude, ChatGPT, Codex, Grok, any MCP client) | In Paseo's own app or CLI |
| Coding agents supported | Any headless agent CLI (claude -p, codex exec and others), started as a detached job |
Claude Code, Codex, Copilot, OpenCode, Pi, Antigravity, Muse Code, more via ACP |
| Parallel agents in worktrees, diff review, PR flow | ✗ No workspace app; use your AI client and your own git workflow | ✓ Worktree per agent, diffs, pull requests, terminals and a browser |
| Mobile app | ✗ No app of its own; works from any AI client that can use MCP | ✓ Native iOS and Android, with voice input |
| Shell commands and detached jobs on many machines, from an outside AI client | ✓ Commands and detached jobs; screenshots on desktop macOS and Windows; file transfer on Pro | Terminals inside agent workspaces; MCP, CLI and SDK automation of agent sessions; arbitrary commands for an outside AI client not documented |
| Remote access | ✓ Agent dials out over one WebSocket; no inbound port, no SSH, no VPN | Optional end-to-end encrypted relay, direct local network, your own tunnel, or an exposed daemon port |
| Self-hosting and licence | Hosted relay; no self-hosted option | ✓ Daemon runs on your machines, optional relay; Apache 2.0 |
| Price | Free for up to 10 machines; see pricing | Free; funded by sponsorship and Paseo Hub |
"not documented" means exactly that: we looked at the sources listed below (2026-11-12) and found no statement either way. It is not a claim that the feature is absent, and we make no claims at all about the other project's security, reliability or roadmap.
Where AI Commander is different, concretely
- No inbound anything. The agent opens one outbound WebSocket to the relay. No port forwarding, no exposed SSH, no VPN — a machine behind NAT or CGNAT works the same as one with a public IP.
- Your chat is the interface. There is no app to install on your phone or desk: Claude, ChatGPT, Codex, Grok or any MCP client calls AI Commander's tools and names the machine, so "on gpu-rig" and "on nas" work in the same conversation.
- Short work and long work are different tools. A command is capped at 1 hour of wall clock, which is a hard kill, and 1 MiB of output, which truncates the reply. Anything longer goes to a detached job, which keeps running after the call returns, after the client disconnects and after the chat ends; you read its logs by job id from a new conversation.
- Jobs can survive an agent restart, with a condition. On Linux each job is launched into its own transient
systemdscope, outside the agent service's control group, so restarting or upgrading the agent leaves it running — that needs a systemd host with the agent running as root, and without both a restart still ends running jobs. - GPUs are first-class. Each machine reports its NVIDIA cards with model, VRAM and utilisation, and passing
gpu_indexreserves one card exclusively; a second job wanting it is refused withgpu_busy. - Commands are not kept. Command text and stdout/stderr pass through the relay's memory and are discarded — never written to logs or a database. See the privacy policy.
Price
AI Commander's Free plan is AI access to your computers, up to 10 saved machines.
Pro is $49 per month and covers every saved machine up to a technical 100-machine
ceiling*, plus file transfer with
remote_pull and remote_push. See our pricing page for the full breakdown.
On the Paseo side, checked 2026-11-12: Paseo says it is free and open source under Apache 2.0, that you need agent providers installed with your own credentials, and that the project is funded through sponsorship and Paseo Hub.
* 100 machines is a technical ceiling, not a policy limit. Need more? Get in touch — we'll sort it out.
When NOT to use AI Commander
- You want a dedicated app for parallel coding agents with a worktree per agent, diffs and pull requests.
- You want a native iOS or Android app to steer those agents; AI Commander has no app of its own.
- You want voice input for your coding agents.
- You need a self-hosted, open-source tool under Apache 2.0; AI Commander's relay is hosted and has no self-hosted option.
- You want to script agent sessions through a CLI, MCP or a TypeScript SDK across providers.
When AI Commander fits
- You want to stay in Claude, ChatGPT, Codex or Grok and have it act on your machines, not switch to another app.
- There is more than one machine — a Windows PC, a Mac, a Linux server, a NAS, a GPU box — and the box may be headless or behind NAT.
- The work is long and must not die with the chat: a build, a training run, or Claude Code or Codex itself started as a detached job.
- You need plain commands and jobs on those machines, not only coding-agent sessions — plus screenshots of desktop macOS and Windows machines, and file transfer on Pro.
FAQ
Sources & verification
Paseo facts on this page were read from its own site and repository on 2026-11-12:
- "The agentic development environment", desktop, mobile, remote machines, Apache 2.0, sponsorship and Paseo Hub, FAQ (free and open source, own credentials, encrypted relay or own tunnel, daemon can run headless, providers run as you normally run them, automation via MCP, CLI or TypeScript SDK)paseo.sh
- Supported agents, worktree per agent, editor, terminals, diffs, pull requests and browser, native iOS and Android, voice and
npm install -g @getpaseo/clifor serversgithub.com/getpaseo/paseo
AI Commander facts trace to llms.txt, the docs, the privacy policy and the pricing page.
Let Claude Code and Codex work on the machines you already have
Install the agent on a PC, a Mac, a server or a GPU box and connect Claude, ChatGPT, Codex or Grok. Free for up to 10 machines.