Paseo: A Remote Control for Your Coding Agents

Your agent is still working. You’re in bed.

You close your laptop and leave. Claude Code is still running back home, chewing on a refactor that took an hour to set up.

That night, on the couch, you want to check how far it got. You can’t. The agent lives on your machine, and when you’re not in front of it, you’re blind. Nudge it forward? Forget it.

That’s not the agent’s fault. There’s a wall between you and it. Paseo removes the wall: agents keep running on your computer, and your phone, browser, and terminal become remotes.

It doesn’t touch your current setup

First, the worry: this won’t replace what you’re using.

Paseo launches each provider’s own CLI. Claude Code runs its own binary, Codex runs Codex’s. Your subscription, your skills, your config, your MCP servers stay exactly as they are. It never proxies model calls, your keys don’t flow through it, and your code stays put.

What it adds is a small background process (a daemon) on your machine that manages the agents. Phone, web, desktop, and CLI are all doors into that process.

The list of supported agents is broad: Claude Code, Codex, OpenCode, Cursor, Pi, and 34+ more, all wired the same way.

What you can do with it

The mobile app is native, not a webview wrapper. After setup, three things cover most of what you’ll do:

• Check status. How far the agent got, what it’s stuck on, whether it finished. Live.
• Keep talking. Remember something mid-walk and send it: “Add more context to this bug’s logs.” It keeps working.
• Read diffs. Review what changed and where, straight from the phone.

Anything the app can click, the terminal can script. paseo ls lists running agents, paseo attach streams output, paseo send adds a follow-up. If the app can start it, the CLI can automate it.

How it reaches your computer

Two setups.

On the same network, a direct connection, no relay involved. Away from home, it goes through an end-to-end encrypted relay. Paseo runs the relay but can’t read your traffic.

Or skip that relay entirely. Tailscale, a Cloudflare tunnel, or just exposing a port all work.

More than one agent, without chaos

Ever held back two agents touching the same repo, worried they’d trample each other?

Paseo isolates them with git worktrees. Each agent lives on its own branch, one fixing feature-a, another feature-b, even their dev servers on separate ports, each reachable through URLs like web.fix-auth.my-app.localhost. Review both diffs in the app. Nothing overwrites.

Then the version where someone tags you

Hub is an optional piece on top of Paseo. Connect GitHub, Slack, or Discord, and “someone tagged you” can start an agent on your machine.

Setup looks like this: drop a .paseo/hub.yml in your repo, spelling out who is allowed to trigger (a from_users allowlist), which machine to run on, which agent, and a max duration. Push, and it’s live. From then on, someone tags you and the agent spins up.

Security corners are covered. Only people on your allowlist can trigger anything. GitHub-triggered steps get a scoped credential that works for that one repo.

The boundary is just as clear: if the machine running your agents is powered off, the event is recorded as failed on the spot. Nothing queues. Hub makes no retry promises.

When to say no

It’s not a hosted service, not an IDE, not a model marketplace. The reasoning happens in the CLI you already use, your keys stay yours, and Paseo just dispatches.

So the test is simple. If you only ever work in front of your desk, the standalone CLI is plenty, and Hub is overkill. If you want to check progress from your phone, let a coworker tag an issue and watch work happen, or slot agents into a CI-shaped flow, Paseo is worth it.

Installing is one line: npm install -g @getpaseo/cli && paseo. The rest you’ll know after you try it.