39 releases 39 tags

RSS feed

• v0.2.0-rc.15 c6ea84695f

  Compare

  v0.2.0-rc.15 Pre-release

  Cody Bryant released this 2026-08-13 22:09:02 -07:00 | 116 commits to main since this release

  Thanks to @CryingPecan, whose LitterHopper robot
  on ESP 1.4.4 is behind much of what's below. Protocol detail lives in
  docs/devices/litter-robot-4/.

  Added

  • The CLI remembers your robots. provision saves the serial, broker and CA
    under ~/.whiskerless, so every later command runs bare: whiskerless state
    instead of whiskerless state --serial LR4Cxxxxxx --host … --ca …. robots
    lists what a machine knows, use picks the default when you own more than one,
    and forget drops the saved details without touching the robot. Every flag
    still works as an override, so nothing that scripts today stops working.

  • Another robot inherits the setup already in use — broker, CA and WiFi network
    are offered at each prompt, so only the serial and the WiFi password (which is
    deliberately never stored) have to be typed. Each field is
    judged on its own — robots that share a broker but sit on different networks are
    still offered the broker — and where the saved robots disagree, the default
    robot's value is offered instead, shown at the prompt (the CA, which cannot be
    shown, is labelled with whose it is).

  • Nothing secret is saved. The broker password is supplied per run
    (WHISKERLESS_PASSWORD, or --password if you don't mind shell history), the
    WiFi passphrase is never kept, and the robot's factory certificate and key are
    neither read nor written. Saved files are owner-only (0600) on POSIX; on
    Windows, which has no mode bits, the user profile's own ACLs are the boundary.

  • whiskerless --version, and a bare whiskerless now says what the tool is
    and which robots are set up, instead of an argparse usage error.

  • The CLI shows signs of life. The BLE scan — the stretch a first-time user
    stares at with nothing moving — draws a live spinner with elapsed time (a
    heartbeat line when piped, so logs show liveness too). monitor and state
    gain color in a terminal, never in a pipe, with NO_COLOR and TERM=dumb
    honored. The empty-cycle and power prompts open with a high-visibility banner,
    so the one question that cannot be un-answered is not read at scroll speed.
    Stdlib only — no styling dependency lands in any install channel.

  • The README walks the whole journey: what using it looks like before any
    install, the physical prerequisites gathered in one place (including which
    label line is the serial), per-platform installs including Homebrew, everyday
    use, upgrading, the release-candidate channel, and uninstalling.

  • A damaged profile is visible and removable. robots lists an unreadable
    profile as such instead of silently hiding it, forget removes one even when
    it no longer loads, and use refuses to make one the default — a corrupt
    entry was previously both invisible and impossible to clear from the CLI.

  • The clean cycle and reset buttons are back, and this time they work: they
    synthesise the same button press the panel sends. Proven on ESP 1.1.75 and 1.4.4.

  • Empty cycle and Power buttons, disabled by default and named (danger) — an
    empty cycle dumps the globe into the drawer, and Power takes the robot off the
    network. The CLI gains empty-cycle and power, which prompt first.

  • Pet weight actually works. Recent rc builds doubled the reading; weights now
    match the household scale (raw ÷ 100, the cloud's own units).

  • LitterHopper support: connected, fill gauge, and an out-of-litter alert the
    firmware itself never raises. The hopper entities switch themselves on the first
    time a hopper reports.

  • Litter level as a percentage. It calibrates itself over time; one button press
    with the globe filled to the line pins it immediately.

  • New entities: last cat visit, last visit duration, waste drawer last moved,
    and panel brightness for bright and dark rooms.

  • Activity-derived entities survive a restart instead of reading unknown until
    the next cat visit.

  • New install channels: brew install sisyphusmd/tap/whiskerless, .deb and
    .rpm for amd64 and arm64, and standalone Linux binaries for both. None of them
    need a system Python — the audience is someone provisioning a robot from a laptop
    that has none.

  • Excess weight detection. The robot refuses to cycle while it thinks something
    is sitting on the scale, raises the condition itself after 30 minutes, and shows it
    on the panel — but says nothing about it over MQTT. One robot here sat like that for
    over two hours after a bonnet was reseated slightly off, with its clean cycle stuck
    the whole time and nothing on the dashboard to explain it. Now a sensor. Pressing
    Reset zeroes the scale and clears it.

  Removed

  • LitterRobot4Client (and WhiskerlessAuthError, which only it raised).
    It had no consumers — the CLI drives LitterRobot4Link, and Home Assistant
    rides HA's own MQTT — while duplicating the write-verify-retry logic a third
    time and claiming, wrongly, to be the integration's client. If a daemon ever
    needs a supervised push client, git history has it, and building it on the
    derived-state library planned in the backlog will beat resurrecting it.

  Fixed

  • The hopper level no longer reads unknown for days after a restart on
    robots whose hopper was proven before the gauge was persisted. The reading
    lived only in the raw sensor's restore cache, and the carry that rescues it
    ran solely inside the once-per-revision upgrade sweep — so an install already
    at the current revision restarted into an unknown level beside a raw gauge
    showing a real number, until the next dispense. The carry now runs at every
    startup (and still refuses implausible cached values).
  • brew install sisyphusmd/tap/whiskerless works again. The formula pinned
    bleak 3.x, whose build backend (uv_build) Homebrew cannot build from source, so
    every install failed after the tap had already published. The formula closure
    now pins bleak 2.x (the library itself is unaffected), and every release now
    installs the rendered formula from the local sdist in a linuxbrew container
    before the tap publishes — the failure that shipped (a build backend no
    platform could build) can no longer pass silently. A macOS-only resource
    breakage could still; the closure's platform split is small and the .pkg CI
    covers the macOS binary itself.
  • A mistyped path is now a sentence, not a stack trace. provision answered a
    CA path of ~/.whiskerless/ca.crt with a FileNotFoundError traceback and
    PyInstaller's "Failed to execute script" — and ~ was the reason: the path is
    typed at a prompt inside the program, so the shell never expands it. ~ is now
    expanded everywhere the CLI takes a path, and file and broker errors — including
    a broker that drops mid-session — print one line and exit. (BLE-stack failures
    during provisioning can still trace back; translating them at the library
    boundary is backlog #64.) --debug (or WHISKERLESS_DEBUG=1) still gives the traceback
    for a bug report.
  • provision checks each answer as you give it. The CA was read after the
    serial, broker, SSID and WiFi password had all been collected, so a typo in the
    third answer threw away all five — including a password typed blind. A bad answer
    now costs one line.
  • The serial validator no longer accepts the model number printed beside it on
    the same label. A wrong serial provisions cleanly and then never appears on the
    broker, with no error to see.
  • --dry-run no longer describes writes it never performed. It printed
    "CERT_AWS_ROOT_CERT written" and "APPLY_CONFIG committed" in the past tense, so the
    only thing distinguishing a simulation from a real run was the final line. It now
    says up front what is real (the connect, discovery and reads) and what is not.
  • The hopper entities no longer disappear from a robot that has one. An upgrade
    sweep retired hopper detections recorded from the link register, which is right —
    that register proves nothing. But it also cleared them on robots whose hopper was
    genuinely proven, and the replacement evidence is a dispense, which only happens
    when the litter is actually low. A well-fed robot could go weeks without one. The
    sweep now recognises a previously recorded fill gauge as the proof it is, since
    only a dispense can produce that number.
  • Handling the robot no longer shows up as a cat visit. A Reset press closes a
    visit on the same register a cat does; two of them were published as genuine
    four-minute and three-minute visits. A visit now needs something to have actually
    broken the beam, which a hand on the bonnet does not.
  • The globe motor fault sensor could sit at off through a real fault. It read
    the state document, and the state document does not carry the fault: a robot raised
    one on its activity stream, held it for fifty minutes and cleared it, while
    globeMotorFaultStatus reported no fault in every single state document it published
    in that window. The sensor now watches both channels, and either one raising a fault
    is a fault.
  • The LitterHopper is now detected by watching it deliver litter. The link
    register 0x57 looked like the answer, but a narrated session produced healthy
    readings from it with the hopper sitting on a bench, and its "disconnected" code
    from merely opening the hopper's drawer to refill it — so a refill could park the
    hopper sensor on disconnected with nothing on the wire to ever clear it. Nothing
    is derived from that register any more. Hopper reports connected once litter
    has actually been dispensed and never reports disconnected, because no signal for
    that exists.
  • A robot that dispenses but rarely reports its link no longer loses its hopper
    telemetry. Requiring 0x57 to corroborate a dispense discarded every fill
    reading on such a robot and left its four hopper entities disabled indefinitely.
  • The hopper level survives a restart. Dispensing only happens when the litter
    bed is actually low, so a well-fed robot can go days without one; the last gauge
    is now remembered instead of the level reading unknown until it next runs low.
  • Event sensors now appear only once their fact has actually been reported. Pet
    weight, last cat visit, and waste drawer last moved start hidden and switch on at
    their first real report — some firmware never emits the drawer event or a weight,
    and those sensors read unknown forever there. A one-time sweep on upgrade applies
    the same standard to existing installs: sensors whose values were real stay, hopper
    and visit-duration detections re-prove themselves at the next report (a robot with
    a real hopper re-enables within one visit), and phantom entities disappear.
  • Last cat visit now updates on every robot. It stamps from the occupancy signal
    itself, not only from weight events — one robot has visits but has never weighed
    anything, and its visit sensor stayed empty.
  • Cat detection no longer mistakes weight on the scale for a cat. The occupancy
    field is two bits — one for what the robot can see, one for what it can feel — and
    a bonnet reseated slightly off held the second bit for over two hours, which the
    robot itself reports as an "excess weight" fault. Occupancy and litter calibration
    now use the bit that tracks the animal.
  • Fewer unknowns while calibration settles. The litter calibration reference
    shows the built-in default (marked source: default) instead of unknown, and the
    hopper level shows a labelled estimate until the empty floor has actually been
    learned.
  • The weekday sleep schedule now arms every day. It is a per-day bitmask, not a
    switch, and turning it on armed Sunday alone — so it looked fine if you tested on
    a Sunday and did nothing all week.
  • The panel sleep schedule can actually be set. The sleep and wake time entities
    wrote a register the firmware only computes, so they never did anything.
  • Litter readings are suppressed while the globe is not level. Mid-cycle the
    sensors read the globe rather than the litter, and that was published as a level.
  • robotStatus 10 is the clean cycle on every firmware; the old map called it a
    cat pause. The boot cycle and the filter-change wizard are mapped too — both move
    the globe, and while unmapped their readings published as real litter levels.
  • whiskerless set night-light-mode auto works. Every spelling the command
    accepts used to crash — and it is the command in the README quickstart.
  • The declared Home Assistant minimum was wrong (2025.2.0), so a user on 2025.2
    could install this and watch it fail. It is 2025.3.0.
  • Every settings write is verified by reading it back, a multi-register write no
    longer loses one of its parts, and broker failures are reported rather than raised
    as a traceback.
  • The hopper no longer drops to unknown on a link code that is not a disconnect, and
    one dispense can no longer prove an empty hopper on its own.
  • The out-of-litter alert judges against your robot's own learned floor, not a
    fixed gauge threshold taken from one unit. Floors differ per robot — one unit's
    stocked readings sit below another's empty flatline — so the fixed cutoff could
    cry empty while litter still flowed. Until the floor is confirmed (which the
    first genuine empty itself teaches), the alert stays quietly off.
  • A restored "excess weight" alarm clears when the robot says the pan is clear.
    It used to survive the clear and re-fire at second zero of every later cat visit
    for the rest of the session.
  • A restored globe-motor fault can finally turn off after a missed clear. If
    Home Assistant was down when the fault cleared, the alarm re-restored itself on
    every restart forever; a clean cycle completing without a fault event now clears
    it, since a faulting cycle would have raised one.
  • A detection re-sweep no longer disables entities you enabled yourself.
    Retiring old detection evidence used to revert a deliberately enabled pet-weight
    or hopper entity on every evidence-standard revision.

  Changed

  • Breaking: binary_sensor.<robot>_waste_drawer_removed is replaced by
    sensor.<robot>_waste_drawer_last_moved. The robot reports that the drawer moved
    and never which way — nine codes turned up across removals and insertions alike.
  • Breaking: clean cycle wait time is a number (3–30 minutes), not a select. Move
    automations from select.<robot>_clean_cycle_wait_time to
    number.<robot>_clean_cycle_wait_time.
  • Breaking (library): Hazard.MOTOR, MotorCommandError and allow_motor are
    gone. A written press is the same event as a physical one, so the robot's own
    interlocks apply either way. Power still requires allow_dangerous.
  • Last visit duration is not reported by every robot, so it ships disabled and
    switches on the first time yours reports one, rather than reading unknown for the
    life of a robot that never will. This was thought to be a firmware split; it is not.
    Two robots on the same ESP build sit either side of it.
  • The raw binaries and the macOS installer now carry the version in their
    filename (whiskerless-<version>-linux-x86_64,
    whiskerless-<version>-macos-arm64.pkg) — a file sitting in Downloads now
    says which release it came from, pairing with --version for the running one.
    The naming scheme is documented in packaging/README.md.
  • Breaking: switch.<robot>_panel_sleep_mode is now
    binary_sensor.<robot>_panel_sleep_mode. The firmware computes that register
    from the weekday schedule and refuses direct writes, so the switch was a control
    that could only time out and error. The weekday sleep schedule switch and the
    sleep/wake time entities are the writable path.

  Downloads

  • Source code (ZIP)
    0 downloads
  • Source code (TAR.GZ)
    0 downloads

  ---

  • whiskerless-0.2.0-rc.15-linux-arm64
    1 download · 2026-08-13 22:34:58 -07:00 · 26 MiB
  • whiskerless-0.2.0-rc.15-linux-x86_64
    1 download · 2026-08-13 22:34:58 -07:00 · 26 MiB
  • whiskerless-0.2.0-rc.15-macos-arm64.pkg
    1 download · 2026-08-13 22:12:10 -07:00 · 11 MiB
  • whiskerless-0.2.0-rc.15-macos-x86_64.pkg
    1 download · 2026-08-13 22:12:12 -07:00 · 12 MiB
  • whiskerless-0.2.0~rc.15.aarch64.rpm
    0 downloads · 2026-08-13 22:34:58 -07:00 · 26 MiB
  • whiskerless-0.2.0~rc.15.x86_64.rpm
    0 downloads · 2026-08-13 22:34:58 -07:00 · 26 MiB
  • whiskerless_0.2.0~rc.15_amd64.deb
    0 downloads · 2026-08-13 22:34:58 -07:00 · 26 MiB
  • whiskerless_0.2.0~rc.15_arm64.deb
    0 downloads · 2026-08-13 22:34:58 -07:00 · 26 MiB