toWidget `io.lotus.set_screenshare { on?: boolean }` (omit to toggle)
starts/stops sharing via the view model instead of the host clicking EC's
hidden button. The host sends it with Capability Delegation
(`delegate: "display-capture"`) where supported, so getDisplayMedia keeps
the user's click on engines that require it. `controls_state` now carries
`screenshareAction: true` so the host only uses it on a fork that has it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The host had only a boolean `speaking` for yourself, throttled to 500 ms, so
nothing on the call bar showed that the mic was actually picking you up.
- lotusMicLevel.ts: one shared local mic sampler (a clone of the published
track, ~10 Hz RMS, while a mic is published, muted or not), shared per
call view model so there is one AudioContext.
- MicLevelQuantizer: 0–3 bars (≈ −46/−36/−26 dBFS), rising at once and
falling one bar per sample so it doesn't flicker between words.
- fromWidget io.lotus.mic_level { bars } only when the value changes; 0 while
muted or with no mic. Opt-in with the host state stream (lotusCallState=1).
- "Talking while muted" (#37) now reads the same sampler instead of running
its own; same gate and behaviour.
Verified in a local call with a fake-tone mic: the host receives 0–3 as the
tone rises/falls (about 4 messages/s while it changes), 0 on mute and levels
again on unmute; muted + tone still raises speakingWhileMuted and the host's
"You're muted" notice. Lotus/room suites pass on Node 22 (148 tests).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The per-participant volume slider lived only in the tile's view model, so it
reset to 100 % on the next call, a reload, or when that person reconnected.
- lotus/lotusVolumeMemory.ts: per-Matrix-user volume in localStorage
(`lotus-per-user-volume`), keyed by user id so it follows them across
devices; 1 (100 %) forgets the entry; at most 50 people, least recently set
dropped first; junk in storage ignored. Local only, never synced.
- VolumeControls take an optional initialVolume and onCommit; remote user
media seeds from and saves to the memory. Screenshare audio stays per-share.
- commitPlaybackVolume now takes the slider's committed value. With the
keyboard the slider commits before its last change arrives, so the
committed (and remembered) volume was one step behind, and returning to
100 % with End never registered.
Verified in a two-person local call: Alice sets Bob to 70 % with the
keyboard → audio element 0.70, stored 0.7; leave + rejoin → slider and audio
0.70; End → entry removed. Unit tests: memory module (LRU, reset, junk),
remembered volume restore + commit-with-value; state/tile/room/lotus suites
all pass on Node 22 (410 tests).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The Lotus host drove these by clicking EC's hidden controls in our DOM and
read screenshare/layout state from it, which only works while the frame is
same-origin and breaks whenever a data-testid changes. New actions:
- toWidget io.lotus.set_layout { layout: "grid" | "spotlight" } → the layout
switch view model's setLayout.
- toWidget io.lotus.open_settings { open?: boolean } → setSettingsOpen
(toggle when `open` is omitted).
- toWidget io.lotus.toggle_reactions → a window event the reactions button
listens for.
- fromWidget io.lotus.controls_state { screensharing, layout } on change; its
arrival tells the host the fork supports the actions above.
Screensharing stays host-driven for now: getDisplayMedia needs the user's
click to reach this frame (Capability Delegation), which a widget message
doesn't carry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
* Give the microphone level its own observable
* Add a microphone level meter
* Show speakers and microphones in the quick audio menu
* Wire speaker selection through the call footer
* Add stories for the meter and the device menu
* Add end-to-end specs for the quick audio menu
* Keep the footer while a menu opened from it is open
* Use the shared audio capture stub in the lobby test
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Say what earns a comment its place
- Add a comments section to the code style: no comment by default, the reason
only, and the kinds that go stale.
- Point to it from the PR checklist in AGENTS.md.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
"Failed to connect to Livekit server (Reason: ServerUnreachable)" left a
user with nothing to try; they retried for a long time and gave up.
- Title "Couldn't connect to voice". The message depends on the reason:
unreachable/timeout-style errors explain that voice needs its own live
connection, which VPNs, antivirus web shields and some networks block,
even when chat works, and say what to try; NotAllowed says the call may
be full or access changed.
- The raw reason, status and message move under Technical details.
- LivekitConnectionError and PeerConnectionTimeoutError now get a
Try again button (the existing reconnect path), shown as primary: the
Lotus theme renders `secondary` as dark-on-dark.
Verified in the Lotus client against a local LiveKit with the signal
socket and validate request blocked: the dialog appears after ~20 s;
unblocking and pressing Try again joins with the mic published.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Fixes the following warning:
[warn] (!) Your Vite config uses features that are unsupported by `configLoader: 'native'`, which is planned to become the default in a future major version of Vite:
- import "./vite.config" without a file extension (vitest.config.ts:3:35). Add the file extension
Set `VITE_CONFIG_NATIVE_IGNORE_WARNING=true` to suppress this warning.
As of the removal of 'Legacy' mode, there can no longer be a difference between the transport you advertise in your membership and the transport you publish media on, so the local transport code can be simplified considerably. For instance, the function which gets the local transport can simply return a promise rather than being reactive.
I kept the local transport as an Observable in other modules so that they could easily be tested with existing marble tests.
LiveKit's own ended handler runs before ours and unpublishes the share;
a listener removed during that dispatch never fires. Decide from the
track's readyState on LocalTrackUnpublished as well. Verified end to end:
closing the shared source now toasts in the host.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Watches the local screenshare publication and tells the host, once per
share each: the MediaStreamTrack ended (window closed), it has been muted
(no frames — minimised/occluded window) for 15 s, or 30 min of sharing
with nobody else in the call. Detection only; the host renders the
notices. LiveKit already unpublishes an ended share, so nothing lingers.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The spotlight screenshare tile already shows a speaker glyph only when the
share publishes audio. Two gaps: a publication the sharer had muted still
counted as audio, and the host's screenshare-audio mute (io.lotus.set_deafen
→ renderer muted) wasn't reflected — the glyph stayed "on" while nothing
played. audioEnabled$ now tracks publication presence AND !isMuted via
observeParticipantMedia; the glyph shows crossed when either the tile's
own mute or the Lotus mute is active.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The embed hides Element Call's footer, so its own output picker is out of
reach; the host's call bar now sends io.lotus.set_audio_output { deviceId }
and the fork applies it via mediaDevices.audioOutput.select(), answering
with io.lotus.audio_output_state { deviceId } on every change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Tracks the local participant's LiveKit ConnectionQuality (time spent
poor/lost) and Reconnecting events for the life of the in-call view and
sends one io.lotus.call_summary { durationMs, reconnects, poorMs, verdict }
to the host on the first SFU disconnect or on teardown. Nothing is stored
or sent anywhere else. CallQualityTracker is pure and unit-tested.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
While the mic is muted LiveKit keeps the capture alive and merely disables the
published MediaStreamTrack, so VAD says nothing. lotusMutedSpeech taps a CLONE
of the published track (post-denoise when the in-source processor is active),
samples RMS at 10 Hz through an AnalyserNode and runs a hysteresis gate (300 ms
of voice on, 800 ms of quiet off, threshold ≈ −36 dBFS). The tap only exists
while a muted mic publication exists; the flag is set on the local entry of
io.lotus.call_state only and is never sent to other participants. Gate is
unit-tested; verified end-to-end: muted with a tone mic → true within 2.5 s,
silence → false, unmute → false.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
mapAuthErrorToUserFriendlyError turned every non-whitelisted error back into
FailToGetOpenIdToken, so lotus.5's refusal reason still surfaced as the
generic page (caught end-to-end with a routed 403 on /sfu/get). Pass it
through like the other user-facing auth errors. Unit-tested.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
When the voice-limit guard (or any JWT service) answers 403 with a reason —
"This voice channel is full.", "You don't have permission to …" — EC wrapped it
in FailToGetOpenIdToken and the user saw the generic error page with
OPEN_ID_ERROR. A 403 MatrixError with a non-empty `error` now throws
SFUTokenRefusedError ("Can't join this call" + the server's sentence), so the
reason is the description. Other failures are unchanged. Unit-tested.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Lotus CIandClaude Opus 5.5
renovate[bot]
Johannes Marbach
fkwpandClaude Opus 5
Robin
Half-Shot
Timo K.