crosspoint

A read-only terminal UI for inspecting Dante and AES67 audio networks over SSH.

Unofficial and read-only. Not affiliated with, endorsed by, or supported by Audinate. Dante is a trademark of its respective owner; this project is not a Dante product and ships no vendor branding.

A crosspoint is one cell in a routing matrix. The matrix is the centrepiece of this tool.

What it is for

Dante Controller is a desktop GUI. When the rack is at the far end of a building and all you have is a shell, there is no equivalent. crosspoint shows device state, the subscription matrix and clock health in a terminal, and never writes to a device.

Status

Early. This session's build runs against JSON fixtures only.

backend	what it does	state
mock	JSON fixtures, deterministic, zero network	working
pcap	replay a saved capture	not built
live	real mDNS discovery plus ARC queries	not built

Install

uv sync
uv run crosspoint --backend mock

Two console scripts, same entry point: crosspoint and xp.

--iface IFACE      network interface Dante lives on
--backend {mock,pcap,live}
--fixture PATH     JSON fixture or capture file
--ascii            plain ASCII glyphs instead of Nerd Font

The default glyph set needs a Nerd Font. --ascii swaps the whole set for plain consoles and serial terminals.

Driving it

Nothing here is keyboard-only. The mouse does what it does in a desktop GUI:

• Click a cell in the routing grid to see what is wired to what.
• Click a device name down the left of the grid to open or close its channels; click a transmitter name along the top to open or close its columns.
• Click a device in the list or the Devices table to select it.
• Click a column header to sort by it, click again to reverse.
• The wheel scrolls.

A legend sits under the grid, and ? opens a key reference generated from the bindings themselves, so it cannot drift out of date.

j k h l ← → ↑ ↓   move a cell        1 2 3 4  devices / routing / clock / events
g G               top / bottom       Tab      cycle focus
PgUp PgDn ^b ^f   page rows          \        toggle the device list
H L               page columns       r        reload
0 $ Home End      first / last col   ?        help
n N               next / prev problem cell    q  quit
m M               next / prev problem on this device
p                 cycle all / problems / errors
s S               sort by next column / reverse it  (Devices and Clock tabs)
Ctrl-P            jump to a device by name
[ ]               prev / next TX device block
Space             open / close this receiver (row)
t                 open / close this transmitter (column)
E C               open all / close all
Enter             inspect the crosspoint under the cursor
/                 filter, Esc clears

The estate at a glance

The header carries the verdict — 11 bad routes, 1 offline, or all healthy — before anything else on the screen.

The Devices tab is one row per device: state, problem count, clock role, sample rate, latency, channel counts. Sort by any column, because "which device is on the wrong sample rate" and "which device has the most broken subscriptions" are questions about the whole estate, not about one device. The device under the cursor is expanded underneath.

The device list down the left side groups by name prefix once grouping actually groups — CR03-Ceiling-Mic files under CR03 — so 200 devices read as the dozen rooms they live in, each with its own offline and fault tally. Selecting a device anywhere selects it everywhere, and parks the routing grid on it.

The routing matrix at scale

A 200-device network is a 12,800 x 12,800 grid, and it is 99.99% empty — an RX channel subscribes to at most one TX channel, so there are at most 12,800 filled cells in 164 million. Rendering that is not the problem (only the visible window is ever drawn, ~3 ms a frame either way). Crossing it is: 12,800 keypresses per axis.

So device blocks start collapsed. The default view is one row and one column per device, and each cell shows the worst status of every subscription underneath it — which device pairs are unhealthy, on one screen. Open a block with Space (rows) or t (columns) when you want its channels.

The grid is glyphs, so a legend is pinned under it: ◆ connected, ● idle, ▲ warning, ✕ error, · no route. ERROR is a distinct shape as well as a distinct colour, so the display never depends on colour alone.

Navigation is built around jumping rather than scrolling:

• n / N — next / previous problem cell. This is the main way to work a large rack.
• [ / ] — previous / next TX device block.
• H / L, 0 / $ — page and jump along the column axis.
• / — filter. Bare words narrow rows; tx: narrows columns. So rack04 tx:dsp shows rack04's inputs against the dsp's outputs. Filtering both axes off one word would empty the column axis the moment you typed an RX device name.

A status line under the matrix always shows the RX and TX device and channel at the cursor, your position on both axes, and the total problem count — at this size you should never have to guess where you are.

Read only

v1 does not change subscriptions, rename anything, or write configuration. It is meant to be safe to point at a production rack. Anything that would mutate a device is listed in DEFERRED.md and is not implemented.

Protocol source

Protocol knowledge comes from network-audio-controller (netaudio, Unlicense) — its packet dissector and data model are the reference. Nothing about the wire format is guessed here; fields netaudio does not cover are marked TODO in the source and left alone.

Subscription status codes are mirrored from netaudio's SUBSCRIPTION_STATUS_* constants and each is mapped to one of OK / ACTIVE / WARN / ERROR. Note that netaudio 0.2.5 builds its status table at import time from a labels file the published wheel does not ship, so the codes in crosspoint/model.py are frozen from the last revision that hardcoded them (tag v0.0.12) rather than imported.

The live backend, when built, will import netaudio rather than shell out to its CLI. It is an optional extra (crosspoint[live]) because the wheel is platform-specific and carries a native library plus a large dependency tree that mock and pcap do not need.

Theme

"pitwall" — red on black, brutalist. Square borders, no gradients, strict column alignment. Status colours are a separate scale from the theme chrome: OK is deliberately quiet so that healthy cells recede, and ERROR is always paired with a distinct glyph so the display never depends on colour alone.

Tests

uv run pytest
uv run pytest --snapshot-update   # after an intentional visual change

Snapshot tests cover one render per tab against the mock fixture. The fixture contains a broken subscription, a clock slave with a large offset, an offline device and a 64-channel device — the states worth rendering well.

tests/test_scale.py builds a 200-device network and asserts the properties that keep it usable: the default view is one row and column per device, _rebuild cost tracks the subscription count rather than devices squared, n visits only problem cells, a row filter does not empty the column axis, and a frame still renders in well under 50 ms with every fold open (13,000 x 12,800).