View multiple SSH sessions at once — and watch your agents work.

01  /  what you see

Every tab, a live tile — all on one screen.

Toggle the grid icon in the tab bar and every open session becomes a tile at once. The columns grow with the session count — one session fills the window, up to four make two columns, and so on up to a 4×N grid — and the tiles resize to fit, so you never scroll to find a server. Each tile is the real terminal, fed by the session's own live output, in that host's colour theme.

• ▸A connection dot & the host name in every tile header — green when connected, grey when not.
• ▸An "agent" chip with a robot icon appears while an agent is active in that session, so you can tell apart what you are driving from what an agent is.
• ▸The Ember glow outlines a tile the whole time an agent command runs, and for a couple of minutes after it last acted — then it fades on its own.
• ▸An expand button opens any tile as the full active tab and leaves the grid.

02  /  step in

Click a tile and type — you're in that session.

The tiles are read-only mirrors until you click one. Click, and the pointer and keyboard go straight into that session — the same session the agent (or you) already had open, live, with no reconnect and no new shell. Take the keyboard to answer a prompt, correct a command, or just keep working; click another tile to move on. When you are done, expand the one that matters or toggle the grid off to drop back to a single terminal.

03  /  one command, every session

Type once, run it everywhere.

Under the grid sits a send-to-all bar. Type a command and press Enter to run it in every connected session; the ^C and q buttons send those keys raw to all of them, to break out of a top or quit a pager across the fleet. The Snippets picker runs a saved snippet — with its {{variables}} filled in — on every session at once. A snackbar confirms how many sessions it reached, and the bar never logs what you typed. Line up a whole folder of prod hosts and drive them together.

04  /  which sessions appear

Agent sessions from the apps you already use.

A tile lights up for any session an agent opens through the desktop app's local MCP server — terminal_open and terminal_run, driven by Claude Code, Codex, Cursor, Gemini CLI or any MCP client. Those run in your live terminals, so they tile and they glow.

To keep it honest: headless MCP calls (ssh_exec, session_*) and the hosted MCP server that ChatGPT and claude.ai use run outside the app and do not appear as tiles. They are not invisible — every one is in the session recordings and the audit log.

• ▸Off until you turn it on. Agent access is off by default, and letting an agent type into your live shell is a separate toggle again.
• ▸Per-host policy. Set each host to Full, an Allowlist of commands, or Blocked — a blocked host refuses the agent outright.

05  /  rewind it

Watch it live, replay it later.

Recording is on by default and captures output only — never your keystrokes, so passwords typed at a hidden prompt never land in the tape. Every agent command writes an agent ▸ command marker before it runs; on the replay timeline those markers are emerald and your own are amber. Click a marker to jump to it, hover for the label, and scrub the whole run. Behind it all sits an audit log of connections, file operations and agent commands.