The pane framing a linked machine's shell asked you to sign in again. Its tunnel is gated on hanzo.id, hanzo.id refuses to be framed — correctly, a login page inside someone else's document is clickjacking — so the gate's redirect could not run in the frame, and the workspace fell back to opening a second sign-in in a second tab. For the same person, already signed in, on the same identity. Per machine, and again every three hours. The tunnel is now asked directly, with the token this page already reads the registry with, and answers with the session the redirect would have set. What that removes is a whole shape: a pane URL is MINTED for every machine now, never derived, so the workspace no longer knows whether a pane is a sandbox we started or a machine you linked — they differ only in who serves the page and what credential opens it, and neither is a layout's business. The rescue collapses with it. One card, one button, for any credential that stopped working: ask for another. A mint that never lands offers it too, rather than leaving a pane waiting on a frame that was never made.
Hanzo Tabs
Keep tabs on your agents.
Your coding agents work in shells on real machines. Tabs puts every one of those shells in front of you — split, tiled, and reachable from anywhere you can open a browser.
hanzo link # on any machine with a shell
Then open tabs.hanzo.ai.
Why it is shaped this way
There is no backend. Not "a small one" — none. hanzo link publishes a
terminal from the machine over a zero-trust tunnel, already authenticated and
already gated, and the browser frames it directly. A server here would be a third
party to a conversation that has two, plus one more thing to keep running and one
more place a credential could sit.
A shell is a URL. ttyd runs with --url-arg behind a wrapper that names a
tmux session, so ?arg=build is a shell called build and tmux new -A
attaches or creates. One link serves as many independent shells as you ask for.
There is no spawn endpoint because none is needed — and closing a pane loses
nothing, since reopening the name reopens the session with its scrollback.
The layout is a tree, and that is the whole design. A list of rectangles is
the shape that looks easier and then cannot answer the only question that
matters: close a pane, and who gets the space? With a tree it is structural — the
space belongs to the sibling, so closing replaces the parent split with that
sibling and nothing else moves. lib/tiles.ts is that tree and nothing else:
split, close, resize, sweep order, page projection — every one a pure rewrite,
tested without a browser.
Rectangles are derived, and that keeps a shell alive. Nesting panes in the
DOM the way they nest in the tree remounts every <iframe> on any split, and
each terminal reconnects. So geometry() projects the tree onto percentages, the
panes render as one flat list that never reorders, and only their coordinates
change.
A phone pages; it does not tile. 390px cannot hold two terminals and stay legible. Side-by-side splits become swipeable pages, stacked splits stay stacked, and how many share a page is derived from the room — landscape gets one rather than three slivers.
Layout
lib/tiles.ts the layout tree + geometry + the phone projection (pure)
lib/panes.ts what a pane shows: a machine and a shell name (pure)
lib/api.ts the control-plane read, from the browser
components/ the workspace
app/ the marketing page and the workspace page
tests/ 39 tests, no browser required
The terminal client itself lives in hanzoai/cli
(assets/term/) — xterm.js, themed, with a touch key row — because it is served
by the machine, not by this site.
Develop
npm install
npm run dev # http://localhost:3000
npm test # the layout engine and bindings
npm run typecheck
MIT.