Free open-source Alpha / Mac → Windows
Your Mac agent. Working on Windows.
Ask your existing AI agent to run commands, move files and start background jobs on your Windows PC. You authorize the connection. Windows does not need another model.
Confirm on both devices · Manage CLI access · Dedicated Windows administrator account
The idea in 30 seconds
Ask on your Mac. Get the work done on Windows.
Agent Road is the connection tool. Keep using an AI agent with terminal access: it sends commands to your other computer over a connection you authorize, then brings back the output and files.
01 / MAC
Your existing agent
Takes your request · checks the environment · calls the CLI
Tailscale + SSH
02 / WINDOWS
Your other computer
Runs the task · returns results · no second model required
“Check the test environment for my Windows project, run the tests with my approval, and bring back the results.”
An example request, not a live execution receipt. Existing tools are reused where suitable; you approve installation and administrator changes.
See it before you set it up.
A one-minute introduction to Agent Road. The current Alpha connects Mac to Windows; available capabilities depend on the device setup.
Before your first connection
One Mac. One Windows PC. One useful task.
For individual developers with Windows administrator access who are comfortable configuring networking. Alpha is free; fresh devices still need preparation.
New testers need the maintainer-provided bundle: the native Windows installer is not publicly released. If you only have the website, your agent should explain this blocker before attempting setup. Setup guide ↗
Get the test bundle first
Bring one task you want to try.
A free, small-scale Mac-to-Windows Alpha with maintainer assistance. Email your two OS versions, agent and one useful task. The maintainer checks fit before sharing a versioned bundle and instructions.
This opens an email draft; you must send it yourself. No mail app? Copy the address into webmail. There is no automatic enrollment, guaranteed place or promised response time. Do not send credentials, pairing codes or raw logs.
Have a bundle or an enrolled device? Get the agent prompt
Let your agent handle setup
Give this to an agent with terminal access. It must read the official guide before checking and configuring your machine, with you approving the required steps.
- 01
Request a test bundle
Email your setup and one task. After confirming fit, the maintainer supplies a versioned bundle. Do not run pairing commands just to apply.
- 02
Let your agent inspect and set up
Verify the bundle manifest and hashes first. Inspect the environment, approve browser and Windows administrator steps, then follow the bundle instructions.
- 03
Start with one small task
Ask for a read-only command that prints Hello from Windows. Check the returned text and exit code 0, then verify file transfer and background jobs separately.
What changes on Windows? How do I leave?
Setup creates a local AgentRoad administrator account and configures OpenSSH, Tailscale, SSH authorization and firewall rules. Exact changes depend on your environment and bundle. No second account means no second website account.
There is no one-command device detach or complete uninstaller yet. Leaving a trial requires reviewed task shutdown, access withdrawal and verification before cleanup. Website logout is not device detach.
Read the change and withdrawal checklist ↗After signing up, check your confirmation email, including your spam folder.
Useful today
Give your other computer a useful task
For an enrolled Windows PC that passes its capability checks. Inspect the environment first, then run the work you authorize.
- 01
Run a project test. Bring back the result.
Have your agent inspect existing tools, run an approved test command and return its output and exit code.
- 02
Send a script. Retrieve the report.
Transfer individual files with integrity checks. Overwriting is explicit; this is not a full directory sync.
- 03
Leave a job running. Check back later.
Keep the job ID and use it to retrieve status and logs later. Jobs outlive the current SSH connection, but do not survive a reboot.
Boundaries, Not Promises
Consent on both computers
Owners approve account access and pairing. One-use enrollment details are sensitive; do not share them publicly.
Private credentials stay with the controller
The controller stores device SSH keys. Protect its local state and redact logs before sharing.
Verify the connection
Use Tailscale and host-key-pinned SSH. Required checks fail closed rather than silently reducing protections.
You authorize the task
Installations, reboots and security changes require explicit authorization. The CLI can run administrative scripts; it is not a sandbox for untrusted code.
- ×Pre-boot recovery (BIOS / WinRE / BitLocker loops)
- ×Hardware failures (disk, memory, power)
- ×Organization-managed devices
- ×Cross-VPN relays and self-hosted VPS forwarding
Commands. File round trips. Background jobs.
Internal physical-PC and VM trials cover these workflows. Claude and DeepSeek independently used the existing controller. One 24-hour observation recorded 95 successful samples, not an uptime guarantee. An earlier SSH service stop remains unexplained; fresh unfamiliar devices still need external testing.
Check remote output and script exit code
Check Unicode content and integrity on enrolled devices
Reuse the same job ID to retrieve terminal state and logs
Claude and DeepSeek used the existing controller in internal trials
Before you try it
Is this another AI agent?
No. It gives an existing terminal-capable agent remote tools. You keep using your agent; compatibility must be checked for the one you use.
Does Windows need a model or another account?
No second model or website account is needed on Windows, but setup creates a dedicated Windows local administrator account. The Mac controller uses browser account authorization; the Windows owner confirms pairing and administrator changes. Network setup is a separate step.
How is this different from remote desktop?
The validated core is commands, files and background jobs. Mouse, desktop and browser workflows need additional components and a usable session; a connected status does not prove those capabilities.
Which computers are supported today?
The test focus is an Apple Silicon Mac controlling Windows 11 x64 over Tailscale. Mac targets, Linux targets and arbitrary two-computer combinations are outside the current supported scope.
Can I install it now? Is it open source?
Already enrolled devices can be used. New testers should request a bundle through the homepage email and follow its instructions after maintainer confirmation: the native Windows installer is not publicly released. Original source is public on GitHub under Apache-2.0. Public source does not mean a stable installer or unrestricted hosted-service access.
Can it repair a PC that is offline or cannot boot?
Remote diagnosis needs a working authorized channel. Fully unreachable systems, BIOS, BitLocker pre-boot prompts and hardware faults require local intervention.