The sidebar was a flat scan of thirty-one rows; now it reads like a
settings app. Multi-subject categories (Input, Network & Sharing,
Applications, Users & Accounts, Privacy & Security, System) carry an
Appearance-style tab strip above the page, drawn by the shell so the
leaf pages themselves are untouched. The taxonomy lives in one new
file, services/SettingsRoutes.qml; the sidebar, the strip, route
validation, search breadcrumbs, and both generators derive from it.
ShellState.settingsPage still holds leaf ids, so every deep link, IPC
call, and search result keeps working — and now lands on the exact
tab. Dictation moves out of Sound onto its own page under Input, with
a handoff back to Sound for the microphone. The strip scrolls when
System's nine tabs outgrow a tiled window. All 161 contracts pass.
Claude-Session: https://claude.ai/code/session_01Ms2FbjQy31TVf3CEvQhGM8
The Download button on Sound → Dictation ran `panama-dictate download`. There
is no such command: the helper answered {"ok": false, "error":
"unknown-command"} and the button did nothing, every time, for anyone who
tried it. The real command is `setup`, which pulls the speech server image
and downloads the ~490 MB model together -- one trip, by design. And the card
told you the server needed `panama app whisper-vulkan`, an app that exists in
no package or setup/apps entry; the server is a prebuilt image the helper
pulls with podman.
So the card described three things and got all three wrong. It now shows the
real state of both pieces and offers one action -- "Set up dictation" -- wired
to the command that actually fetches them, with progress that names which half
is running (server image, then model). Verified end to end: `panama-dictate
setup` emits the pulling/downloading JSON the service parses.
Claude-Session: https://claude.ai/code/session_01Epx9ZC1gwm81K3jm9x9CKh
Super+D holds the microphone open, releasing it transcribes on the GPU and
types the result wherever the cursor is. Roughly 150ms for a normal utterance
once the model is resident, measured rather than hoped for.
Getting there meant discarding two approaches. Fedora 44 cannot install any
GPU-capable Whisper for Python -- openai-whisper needs a numba that needs an
llvmlite that does not exist for 3.14, and faster-whisper needs a ctranslate2
nobody packaged. The whisper-cpp package IS built with HIP but ships libraries
with no binary and no bindings, and hand-writing ctypes for a large by-value
struct is a segfault waiting for a version bump. So a container, as suggested.
Vulkan rather than ROCm, and upstream's image rather than one built here. ROCm
is seven gigabytes and serves AMD alone; Vulkan compute runs on the AMD, Intel
and NVIDIA machines this config is used on, in a twentieth of the space. The
Vulkan tag already contains whisper-server, so there is no Containerfile to keep
working -- an earlier draft of this commit had one, and it was strictly worse.
Two bugs found by using it rather than by reading it. Whisper describes silence
as the literal text "[BLANK_AUDIO]", and the first working version pasted that
string into the clipboard; a transcription that is nothing but such markers is
now discarded. And the server answers with a line per segment, which typed into
a window is an Enter press -- sending the half-written message, submitting the
form. Whitespace is collapsed to one line.
Neither the image nor the model is installed by ./install. Together they are
over two gigabytes that want the network, and Settings offers both as one
action instead. Nothing starts at login either: whisper-server holds the model
from the moment it starts, so the first press of the key is what brings it up.
The contract pins both text bugs, that the server stays on loopback, and that it
does not start at login. Reverting the [BLANK_AUDIO] guard did not fail it at
first -- the check was still correct, it had simply stopped being called -- so
it now checks the call site too.
Claude-Session: https://claude.ai/code/session_01Q84axqUE5inJhf5Jz9CFy1
Gabriel Brown