Phase B.1 of the native-livekit migration. Ships the command surface and
event bridge behind a cargo feature so the default build stays unaffected
while the JS-SDK path keeps running in production.
Rust side
- livekit 0.7 (default tokio runtime, rustls-tls-native-roots) pulled as
an optional dependency; tokio also optional under the same feature
- Feature `rust-livekit` gates everything — off by default; on via
`cargo build --features rust-livekit`
- src-tauri/src/livekit_bridge.rs: LivekitState mutex, connect /
disconnect / send_data commands, event pump for room_state,
participant_joined, participant_left, data_received
- Mic / camera / screen share commands stubbed with explicit
"not implemented" errors so JS callers fail loudly rather than
silently no-op
JS side
- src/lib/nativeLiveKit.ts exposes a NativeRoom class with the same event
/ method shape the CallContext will need, plus a VITE_USE_RUST_LIVEKIT
flag so the adapter can be swapped once the bridge reaches parity
- isRustLivekitAvailable() gates access at both env + runtime layers
Build impact
- Baseline build unchanged (1s incremental, no new deps pulled)
- Feature build initial: ~10min (libwebrtc download + link)
- Feature build incremental: ~1s
- Binary size with feature: +12-20MB vs baseline
Open questions (documented for the next phase)
- Video-frame rendering bridge remains an upstream gap; livekit-rust
exposes NativeVideoFrame but no stable path to expose that as a
MediaStreamTrack inside the WebView
- Audio-only rust path is realistic near-term; full-rust needs either
upstream video-bridge or a native-overlay render window
Phase A of the crypto/livekit rust-native migration.
Rust side
- dryoc crate (pure-rust libsodium-compat, no C toolchain)
- Tauri commands: crypto_random_bytes, crypto_secretbox_encrypt/decrypt,
crypto_box_keypair, crypto_box_encrypt/decrypt, crypto_box_seal/open,
crypto_pwhash — all bit-compatible with libsodium-wrappers-sumo
- Commands registered via invoke_handler in lib.rs
- All IPC payloads base64-encoded to survive serde_json
JS side
- lib/nativeCryptoOps.ts exposes pwhashArgon2id + randomBytesAsync
plus optional secretbox accelerators for future call-site migration
- Native-first, WASM fallback on error or when VITE_USE_NATIVE_CRYPTO is
false / in browser preview
- Argon2id call-sites migrated: secureFileStore.deriveKey and
deviceBackup.deriveKey (covers vault unlock + backup/recovery flows)
Impact
- Vault unlock: ~1200ms → ~200ms (measured locally, Argon2id moderate)
- Per-message AEAD left on WASM-worker path: IPC overhead ~40µs would
dominate any native speedup below ~100µs/op
- WASM stays installed as graceful fallback so browser-preview builds
keep working and a native failure self-heals at runtime
System tray (desktop)
- tauri tray-icon feature + tray with menu (Öffnen/Ausblenden/Beenden)
- Left-click toggles main window; right-click shows menu
- JS emits tray-unread-update event, Rust mirrors into tooltip +
macOS dock badge via set_badge_label
- ConversationsContext wires totalUnread → tray
Window state persistence
- tauri-plugin-window-state (desktop-only target guard)
- Auto-restore size/position/maximized between restarts
Local SQLite message cache
- tauri-plugin-sql hydration of conversation view on mount
- persistMessages after each refresh, deleteCachedMessage on realtime
DELETE, pruneCache keeps latest 1000 per conversation
- Stores plaintext only (same trust boundary as stronghold device
key; cache never leaves the device, E2EE w.r.t. server unchanged)
Web Worker for decryption
- workers/decrypt.worker.ts runs crypto_secretbox_open_easy + utf-8
decode off the main thread with its own libsodium instance
- lib/decryptWorker.ts is a request/response wrapper with inline
fallback when Worker spawn fails
- shared decryptMessages accepts aeadBatchDelegate so key lookup
stays on the main thread while the AEAD loop offloads
Build fix
- Enable tauri tray-icon feature
- Import Listener + Manager traits, clone tray handle for the
event listener, conditional icon attach
File variety (video/pdf/generic), user status with custom message,
auto online/offline, DND gate for ring + notifications, drag-drop +
paste upload, @mentions in groups, link previews, emoji picker,
crash recovery.
- lib/crashRecovery.ts: window.onerror + unhandledrejection handlers,
dedupe identical messages within 10s, burst-reload after 8 distinct
errors in 30s
- components/CrashToast.tsx: portal stack in bottom-right, max 3 visible,
auto-dismiss after 7s, per-entry close button
- Wired in main.tsx before bootstrap so crypto/i18n errors are captured,
rendered from App.tsx alongside UpdateToast
Covers the async layer ErrorBoundary misses (realtime handlers, stray
Promises, setTimeout callbacks, LiveKit listeners).
Attachments
- Video: native <video controls>, decrypted blob, metadata preload
- PDF: collapsible <object> viewer with download + preview toggle
- Generic: file card with lazy decrypt, mime-to-extension map
- MessageBubble dispatch: audio/image/video/pdf/generic by mime
- Composer accept widened from image/* to any; AttachmentPreview renders
non-images as file cards
User status
- usePeerPresence returns { state, statusMessage } and tracks both fields
via realtime updates
- ConversationHeader subtitle shows custom status_message when peer is
online/idle/dnd (with message set); falls back to localized presence
label; always "Offline" when state === 'offline'
- UserBar: status_message input in dropdown (max 128 chars, save on
blur/Enter), subtitle uses same precedence as peer view
- AuthContext: auto-flip offline → online on session mount; best-effort
offline on beforeunload/pagehide; respects explicit idle/dnd/invisible
- i18n: presence.status_placeholder (DE + EN)
DND
- CallUI: incoming ring suppressed when own presence === 'dnd' (outgoing
rings stay audible — user-initiated)
- CallContext: presenceRef mirrors profile.presenceState, incoming-call
and missed-call OS notifications skipped under DND
- ConversationsContext already gated message tone + notify
Drag/drop + paste
- Page-root drag handlers feed ingestFiles; overlay shown while dragging
- Textarea onPaste extracts clipboard image items
@mentions in groups
- MentionAutocomplete: keyboard-driven dropdown, arrow/enter/tab/esc
- Composer onChange parses trailing @token, auto-open
- renderBodyWithMentions highlights @username tokens in bubbles
Link previews
- Migration 20260421000003_link_previews.sql (cache table, auth read,
service-role write via edge function)
- Edge function og-preview: POST {url}, fetches HTML (6s timeout, 1MB
cap), regex-parses OG/twitter/description, upserts cache
- useLinkPreview hook with in-memory cache + inflight dedup
- LinkPreviewCard rendered from first URL in message body
Emoji picker
- Built-in curated set (~250 emojis) across five categories
- Search by keyword/emoji char/category, recent list persisted in
localStorage
- Trigger button next to + and voice buttons in composer
Soundboard + ringtone + mic pipeline
- Pulled in (user-authored) SoundboardManagerDialog/Panel/Settings,
RingtoneSettings, mic pipeline + hotkeys + storage modules
- AuthContext imports updateOwnProfile for presence auto-flip
Fixes
- AttachmentAudio min-width so bubbles don't collapse to 0px on peer
side
- Focus flicker: visibility/online wake refresh throttled to 30s,
focus listener dropped, loading flag only on first fetch
- Voice messages: MediaRecorder → encrypted attachment, custom waveform
player via OfflineAudioContext, 60s limit + live mic-level meter
- Offline message queue: localStorage outbox, exponential backoff retries,
optimistic pending bubble with retry/discard
- Delivery indicator: message_deliveries table + RLS (reciprocal receipts),
✓ / ✓✓ / ✓✓-blue tick states, group-aware (all members must ack)
- Per-participant volume slider in calls via right-click tile menu,
persisted to localStorage, applied to attached audio elements
- Group call scaling: grid up to 12 tiles with pagination,
active-speaker auto-promotion in fullscreen
- Push notifications scaffolding: service worker, VAPID subscription
registration, notify-push edge function skeleton
- Backup recovery code: 24-char base32 code (~120 bits entropy) as
alternative decrypt path, restore UI with mode toggle
- Admin panel: conversations list, audit log (admin_audit_log table +
admin_log_action RPC), audit entry on user flag toggle
- Search v2: sender filter, attachment-only toggle, date range
- Reactions pop animation (scale 0.4→1.15→1 on count change)
- Message list windowing (150 default, expand via IntersectionObserver)
- Stub cleanup: removed dead ScreenshareStub from CallParticipantTile
Fixes:
- Focus-triggered flicker: dropped window.focus listeners in three spots,
throttled visibilitychange/online wake-refreshes to 30s, keep existing
data visible during background re-syncs (no more spinner on every click)
- Voice attachment audio element collapsed to 0px on peer side — now
forces 280px min-width on bubble
Migrations (push required):
20260421000001_message_deliveries.sql
20260421000002_admin_audit_log.sql
Server TODO:
VAPID keys + notify-push edge function deploy
Speaking ring:
- Switch useActiveSpeakers from LiveKit's smoothed isSpeaking / server-
batched ActiveSpeakersChanged to Web Audio API AnalyserNode on each
participant's raw audio MediaStreamTrack. Poll 50ms, RMS threshold 0.03,
250ms hold. Feels real-time vs the old ~500ms lag
- Defensive syncProbe on every tick so probes catch up if TrackPublished
missed (local mic publish race on join)
- Universal speaking overlay on tile (3px emerald border + inset glow,
z-10) so video mode shows the ring too, not just audio mode
Screen sharing:
- Separate "screen" tile per sharer so the sharer's avatar tile stays
intact with its speaking ring. Tile.id is kind-prefixed (user:xxx /
screen:xxx) so focus tracking distinguishes them
- New ScreenShareDialog (quality preset + fps override + displaySurface
hint) opens on the share button. startScreenShare / stopScreenShare
actions in CallContext replace the one-shot toggle
- ScreenShareViewer: plain CSS-only fullscreen overlay (Tauri WKWebView
doesn't implement requestFullscreen), always `h-full w-full
object-contain`, Esc exits
Camera:
- toggleCamera action in CallContext tracks isCameraEnabled
- VideoStub renders real <video> srcObject for the participant's camera
MediaStreamTrack; local preview is mirrored
- Tile video track resolves to Track.Source.Camera publications of the
LocalParticipant / each RemoteParticipant
- Room listens for TrackMuted / TrackUnmuted and re-publishes remote
state so peers switch to avatar placeholder when a camera is disabled
Deafen:
- New isDeafened state + toggleDeafen action. Sets `muted = true` on all
attached `<audio[data-livekit-track]>` plus mutes fresh ones on attach
via module-level flag
- Broadcast state over the LiveKit data channel
({type:'presence', deafened}) so peers can render the headphones-off
badge. Attributes API not used because the self-hosted server may run
older LiveKit versions
- remoteDeafen: Record<identity, bool> exposed via context, bumped on
DataReceived and re-broadcast on ParticipantConnected
Incoming video call:
- acceptIncoming takes an optional CallKind override so the receiver can
answer a video invite with audio only or promote an audio invite to
video on accept
- IncomingCallPanel shows two accept buttons (audio + video) when the
invite is a video call
Audio devices:
- audioSettings adds inputDeviceId + outputDeviceId, persisted
- CallContext uses them on setMicrophoneEnabled, plus new
setAudioInputDevice / setAudioOutputDevice hot-swap actions.
Output swap applies setSinkId to every attached remote-audio element
since LiveKit's own switchActiveDevice only tracks elements it
attached itself
- SettingsPage "Mikrofon" + "Ausgabegerät" selects with devicechange
listener and a permission-probe button
Fullscreen mode:
- Replaced absolute-positioned speaker + floating thumbnails with a real
flex layout. Default = even grid of all tiles. Clicking a tile flips
to big-speaker + horizontal thumbnail strip. Click focused tile =
back to grid
- Controls overlay pinned bottom; content wrapper has pb-24 so tiles
never sit behind the toolbar
- Grid now uses explicit grid-rows-* so cells get a defined 1fr height
(without it, video intrinsic dimensions blew tiles past the container
bounds on Windows)
UI chips:
- Mic-off badge combines isMuted flag AND
localParticipant.isMicrophoneEnabled, so a user with no mic / denied
permission sees the badge + the toolbar button red even though they
never pressed mute
- Deafen badge on tile chips for local + remote (remote driven by the
data-channel broadcast)
Edit message decrypt:
- handleUpdate in useConversationMessages now refetches the canonical row
via REST after a realtime UPDATE instead of trusting the realtime
payload's bytea encoding. Same pattern as handleInsert — base64 vs
`\x…` hex serialisation varies across supabase/postgrest versions and
was silently producing undecryptable ciphertext for edited messages
on the receiver side
Windows WebView2 background throttling:
- ConversationsContext, useFriendships and useConversationMessages now
listen for visibilitychange / focus / online events and trigger both a
fresh REST refresh and a best-effort channel.subscribe() on wake.
WebView2 aggressively throttles background WebSockets and was dropping
realtime events entirely while the window was minimised, so new
messages and friend acceptances only surfaced after a manual reload
Bump tauri version 0.7.0 -> 0.7.1
Audio device selection:
- audioSettings: persisted inputDeviceId + outputDeviceId
- CallContext: uses stored input deviceId on mic enable, new
setAudioInputDevice / setAudioOutputDevice actions that hot-swap
without reconnect. Output swap applies HTMLMediaElement.setSinkId
to every attached remote-audio element (LiveKit's switchActiveDevice
only tracks elements it attached itself)
- SettingsPage: new "Mikrofon" + "Ausgabegerät" selects with
enumerateDevices, devicechange listener, permission-probe button.
setSinkId-unsupported fallback is messaged but non-blocking
Fullscreen:
- FullscreenCall was absolute inset-0 z-40 which trapped it inside the
<main> pane — sidebar + chat-list stayed visible. Switched to
fixed inset-0 z-[60] so the call overlays the whole window
Discord-style
- ScreenShareViewer fullscreen: CSS-only toggle (native Fullscreen API
unreliable under Tauri WKWebView), portalled to document.body when
active so no ancestor stacking context can clip it. Esc exits
ActiveCallBanner:
- cleanup effect returned early when presence was entirely empty,
leaving the "1 im Raum" fallback stuck after both peers left. Now
schedules dismissLastCall as soon as othersIn.length === 0, with a
3s grace window to absorb presence re-sync flicker
Bump tauri version 0.6.0 -> 0.7.0
Backup / restore flow:
- deviceBackup.ts: v2 format wrapping userId + deviceId + privateKey in
an encrypted JSON payload so restore can re-seed localStorage, vault,
and reattach to the existing server-side device row without provisioning
a new one (conv-key bundles stay valid, no "awaiting key" state)
- shared/auth: restoreDeviceFromServerRecord — verifies session.user.id
matches the backup's userId, confirms the server device row still
exists, then writes the private key into the local secret store
- BackupExportDialog — passphrase + confirm, generates portable string,
copy + download .txt
- DeviceRestore — textarea + passphrase → seeds vault + writes
deviceId cache, treats this install as the original device
- DevicePage now has tabs "Neu einrichten" / "Backup wiederherstellen"
- BackupPromptBanner — post-registration nudge, reads sessionStorage
signal from fresh provisions and persists "never-ask-again" in
localStorage so it stops nagging
- SettingsPage backup section: uses the new dialog; removes the
dangerous in-place key import (restore now lives in the device flow)
Username casing:
- Migration 20260420000002 drops lower() from the handle_new_user trigger
and widens the regex to [A-Za-z0-9_]. profiles.username is citext so
uniqueness + lookups stay case-insensitive regardless of stored casing
- Shared auth: trim() only, no toLowerCase on signup/lookups/search.
ilike handles CI anyway and citext makes client normalisation redundant
- AuthPage regex + input preserve case, FriendsPage search preserves case
- i18n (de+en): updated username_hint / username_invalid / ERR_USERNAME_INVALID
to reflect the new rule
Quick wins:
- React Router v7 future flags (v7_startTransition + v7_relativeSplatPath)
set on BrowserRouter — silences the upgrade warning
- appUpdates.checkForUpdate: swallow benign network/fetch/"could not
fetch valid release JSON" cases silently instead of console spam
- osNotify: persist an "asked" marker in localStorage so the permission
prompt only fires once per install (OS already persists the answer,
but the plugin re-queries loudly otherwise)
Messages:
- Reply-to: hover action, composer chip with cancel, quote bubble inside
the replying message with tap-to-jump + amber highlight ring
- Search: header search button toggles in-conversation search bar with
prev/next + match counter, auto-jump to active match
- Forward: multi-select conversation picker. Attachments are now carried
over: download + decrypt source, re-encrypt under each target conv-key,
re-upload with fresh per-attachment keys, insert new attachment rows
Conversations:
- Archive + mute per member. New migration 20260420000001 adds `archived`
+ `muted_until` on conversation_members. Shared helpers:
setConversationArchived / setConversationMutedUntil / isConversationMuted
- ChatsPage: archive toggle in header with unread badge for archived
bucket, split active/archived lists, muted indicator (BellOff icon,
dimmed unread badge)
- ConversationRowMenu via createPortal (escapes sidebar overflow clip),
forwardRef-based MenuItem so submenu positioning refs survive React 18
- ConversationsContext: suppresses notification sound + OS notif when
target conversation is muted
- Refresh on `profiles UPDATE` realtime so peer avatar / displayName
changes flow to conversation.members without manual refresh
Resilience:
- ErrorBoundary (Discord-style): centred spinner + escalating copy, no
manual reload button. Backoff retry schedule [2s, 4s, 8s, 15s, 30s].
Uses a keyed Fragment (not a div wrapper) so flex h-full chains survive
- App wrapped root + per-route RouteBoundary, conversation-level boundary
- AuthContext: flip `ready` immediately on cached session read; validate
getUser in background so a stalled/offline Supabase doesn't freeze the
app on the loading spinner
Crypto:
- Swap libsodium-wrappers -> libsodium-wrappers-sumo (compact build was
missing crypto_pwhash so Argon2id vault KDF threw, falling back to
plaintext localStorage on every launch)
- Shim d.ts for sumo types (sumo is API superset, no official types ship)
- vite optimizeDeps includes sumo with the "require" condition
- secureFileStore: exists(dir) check before mkdir; surface genuine
permission errors instead of silent catch
Tauri:
- fs scope adds `$APPLOCALDATA` direct entry (not just /**) so the app
data directory itself can be mkdir'd on first launch
Chat layout:
- Skip call_event messages when computing avatar run boundaries so a
regular bubble followed by a call event from the same sender still
shows its avatar
- ConversationsContext: subscribe to profiles UPDATE realtime so
conversation members[].profile picks up peer avatar / displayName changes
without a manual refresh
- ConversationPage: skip call_event messages when computing run boundaries.
Previously a peer bubble followed by a call event from the same sender
was treated as mid-run -> avatar slot collapsed to a placeholder
- DM peer-profile fallback already added in previous commit covers transient
member-lookup misses
Visual:
- Light/dark theme via ThemeContext + CSS vars
- New design tokens (surface/fg/accent/line/etc.) across all pages
- Reusable Avatar component (img + letter fallback) wired into UserBar,
ConversationHeader, ChatsPage, FriendsPage, GroupInfoPanel, IncomingCallPanel
Calls:
- Split call UI: IncomingCallPanel, InCallPanel, CallControls,
CallParticipantTile, ScreenShareViewer
- Active speaker hook (useActiveSpeakers)
- Fix: ActiveCallBanner stayed hidden after hangup while peers in room.
- useCallPresence: bind presence callbacks only when we own subscribe
(Supabase forbids .on() after .subscribe() on shared dedup'd channels)
- useCallPresence: never removeChannel — channel is shared with CallContext
so tearing it down on ConversationHeader unmount killed live tracking
- ActiveCallBanner: lastCallConversationId fallback so banner shows
instantly after hangup, auto-dismiss when room confirmed empty
- Drop unused useAnyActiveCall
Bump tauri version 0.5.0 -> 0.6.0
byGalax