glanced 0.3.4

The lock-screen indicator now survives omarchy update on a pip install.

setup-lock applied the indicator and then had no post-update hook to install, because the hook file lives in packaging/ and a wheel did not carry it. The indicator looked right until the next update reverted it, with nothing to put it back.

Shipped as package data, with the same drift test as the PAM source and the patch directory.

158 tests pass on this tag.

glanced 0.3.3

Setup is one command, and the lock screen finally works from PyPI.

pipx install 'glanced[runtime,gui]'
glancectl setup
omarchy plugin add https://github.com/ayandexyz/omarchy-glance.git --enable

• glancectl setup — the service, the PAM wiring and the lock indicator behind one password prompt.
• The PAM source and the lock patch ship in the package. They were the two steps a pip install could not complete: both failed by opening a terminal that closed before you could read why. A test holds the packaged copies byte-identical to the canonical ones.
• setup-pam now says to install base-devel when there is no compiler, instead of a make traceback.

155 tests pass on this tag.

glanced 0.3.2

glancectl status now reports gui: whether this install can open the enrollment window.

The Omarchy plugin reads it to choose between the window and a terminal, so an install without the gui extra gets an Enroll button that works rather than one that raises ImportError. Answered with find_spec, so nothing imports Qt to work it out.

139 tests pass on this tag.

glanced 0.3.1

glancectl install-service now fetches the models when they are missing, so installing the daemon is two commands:

pipx install 'glanced[runtime]'
glancectl install-service

Previously fetch-model was a separate step, and skipping it produced a service that started and then failed in the journal rather than saying anything useful. --no-fetch keeps the old behaviour.

136 tests pass on this tag.

glanced 0.3.0

The release with attention.sock.

Added

• Attention mode. A third socket, attention.sock, streams head pose (present, yaw, pitch, conf — derived values, never frames or landmarks) to any connected client at --attention-fps (default 8). It uses the landmarker only — no ArcFace, no arming, no enrollment — so it works for anyone with a webcam. The camera is held only while someone is subscribed and is handed over to an unlock scan on request, so an auth request always has priority. glancectl attention prints the stream; glancectl daemon --no-attention turns the socket off. This is what omarchy-attention (a blur shield without screen capture) subscribes to.
• Camera.frames(min_interval) drops frames without decoding them, which is how attention runs the landmarker at 8 fps on a 30 fps camera.

Install

There is no AUR package. From a checkout of this tag:

python -m venv .venv && .venv/bin/pip install -e '.[runtime]'
.venv/bin/glancectl fetch-model
packaging/install.sh            # --no-plugin for the daemon only

126 tests pass on this tag.