Repository navigation
cenno v0.4.0
Includes everything staged under the never-published 0.3.2 bump.
Added
- Every panel now takes a free-form answer. Choice, scale, confirm and
custom-widget panels — and even pure notifications — render a voice-capable
text field below their own control, so you can always ignore the agent's
options and answer in your own words. On Mac, iPhone and Watch. The mic is
offered on every text field, not only when the agent asked for dictation. - Screen context (L1a) — cenno can read the focused window's accessibility
snapshot and expose it to agents through a guarded MCP tool. - Voice-mute — prompts can open silently (
--muted), with an in-panel
mute/unmute toggle; the final mute state is reported in the answer.
Changed
- Voice-out timing — the panel now appears when speech actually starts,
and stopping a voice fades it out (~200 ms) instead of cutting it off;
stop also cancels any in-flight synthesis. - cenno is now pronounced the Italian way ("CHEN-no") when read aloud.
- The About tab shows the live app version instead of a hard-coded one.
Fixed
-
Sloppy agent formatting is repaired on the way in. A body sent with
literal\n/\tescapes rendered those two characters verbatim in the
panel; they now become real whitespace. Windows paths and regexes
(C:\path,\d+) are left alone. -
Tray: "Show pending prompt" is now truthful and functional.
-
The panel no longer times out while you're typing — and never loses your
text. Editing a field now holds the prompt open (the deadline is pushed far
out on every keystroke), and after you stop you get at least a 45-second
think-window before it can expire. The keep-alive reaches the agent side too,
so a late answer still delivers, and in-progress text is saved as a draft as a
safety net. Previously a prompt could close mid-edit and discard what you'd
typed. -
Voice-out respects live config changes. The prompt
panel decided whether to speak using the config loaded at app launch, so
enabling or retuning voice-out in Settings did nothing until a restart —
while "Test voice" (which reads fresh and skips the gate) worked, masking the
bug. The panel now re-reads the voice config fresh for every prompt, so
Settings changes take effect on the next prompt with no restart.