Pluto Door
SSH client with an AI agent and a desktop for every server
122 followers
SSH client with an AI agent and a desktop for every server
122 followers
Meet Pluto Door, the Mac SSH client with an AI agent and a desktop for every server. Ask Ori what's wrong in plain English; it shows every command first, and nothing runs until you approve. Pluto XOS turns any server into a desktop with 15 apps, and Classic keeps a fast terminal, file browser and editor. Works with AWS, Google Cloud and Azure. Your servers, keys and history stay on your Mac. Free for 14 days, then $4.99 a month or $49.99 a year, or pay $59 once and keep every update.
This is the 2nd launch from Pluto Door. View more
Pluto Door
Launched this week
This launch adds Ori, an AI agent that reads your servers, finds what's wrong and explains it in plain English. When something needs to change, you see every command first and nothing runs until you approve; every edited file can be undone. Pluto XOS turns each server into a desktop with 15 apps, Hangar installs software in one click, and scheduled checks notify you before small problems grow. Works with AWS, Google Cloud and Azure. Your servers and keys stay on your Mac. Free for 14 days.
Free Options
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hi Product Hunt! I'm Krishna, the maker of Pluto Door.
I built it after one too many late-night server fixes: logs scrolling past, half-remembered commands, and the worry of running the wrong one on a production box.
Pluto Door started as a fast, native SSH client for the Mac. This launch adds:
Ori, an AI agent that investigates your server and explains what's wrong in plain English. It reads freely, but nothing changes until you approve, and every file it edits can be undone.
Pluto XOS, a desktop for every server with 15 apps: files, processes, logs, services, network, cron and more.
Hangar, one-click installs and updates with the right commands for your distro.
Scheduled checks that notify you before small problems become 2 a.m. problems.
AWS, Google Cloud and Azure accounts in the same app.
Everything stays on your Mac: servers, keys and history are never synced, and Ori uses your own Claude or OpenAI key.
It's free for 14 days, no card needed. I'd love your honest feedback: what would make you trust an AI agent on your production servers?
how does approve work when ori needs sudo, does it ever see the password
@bbueno Good question. Ori never sees your sudo password, and there's no way to give it one. It checks whether your SSH user has passwordless sudo and runs root commands with sudo -n, which fails instead of prompting. If sudo needs a password, Ori tells you, so you can run that step yourself in the Terminal. Even if a command did ask for a password, Ori couldn't see or answer the prompt.
In the default mode, every command that changes your server, sudo or not, waits for you to approve the exact command. Command output is scanned for secrets (password hashes, passwords in config and .env files, keys) and masked before it reaches the AI model. Your SSH credentials stay in the macOS Keychain and are only used to connect.
Showing every command before it runs is the part that matters most to me for an SSH agent — that approval step is what would let me trust it on a real server. I also like the XOS idea of files, processes, logs, and services in one desktop per server instead of juggling tabs. How does Ori narrow things down when the logs are noisy and several things look wrong at once?
@jcbuilds Thanks, that's exactly why approval is built the way it is. Every command shows up with a one-line reason before it runs, and you can approve it, edit it first, or reject it with a note. On noisy logs, Ori follows a fixed order instead of reading everything:
1. Start from the symptom and state. Ori checks whether the service is running, whether the port is open and what systemctl says, before it opens any logs. That alone rules out a lot.
2. Narrow by time, not volume. It reads logs for a window (journalctl -u --since "1 hour ago", the last few hundred lines of the error log) and filters with grep, so old noise never comes into it.
3. Line up timestamps. It compares when errors started with what changed around then: config files modified, packages updated, deploys, disk filling up. "Errors start at 14:33, config edited at 14:30" is usually the answer.
4. Biggest contributor first. When several things look wrong, it ranks them by impact with real numbers (the 38 GB log folder, PID 4912 at 98% CPU) and drills into the top one, instead of chasing every warning.
5. Split it up when it's wide. For bigger problems, Ori can hand separate areas, like the web server, the database and disk, to read-only helper agents that check in parallel and report back.
It ends with a root cause and one to three fixes for you to pick from. It doesn't apply anything on its own; any fix still goes through the same approval step. Ori also keeps short notes on each server (what runs where, which log is noise), so the next investigation starts smarter.