Separate conversation drafts and voice view state, add native settings and restricted CSS themes, implement validated avatar cropping, and refresh dependencies with documented desktop compatibility backports.
6.6 KiB
Desktop Client
The primary VNOX client. Native, fast, lightweight.
Stack
| Layer | Technology |
|---|---|
| Language | Rust |
| UI framework | egui |
| Renderer | wgpu |
| Audio capture/playback | cpal |
| Audio processing | rodio |
| Voice codec | opus (libopus bindings) |
| Networking | quinn (QUIC/UDP), tokio |
| Serialization | prost (Protobuf) |
Why egui
- immediate mode UI: simple to reason about, no complex state trees
- runs on wgpu: same renderer as the rest of the GPU pipeline
- truly cross-platform: one codebase, same behavior on Windows/macOS/Linux
- no runtime dependencies: ships as a single binary
Why wgpu
- modern GPU API (Vulkan / Metal / DX12 / WebGPU backend)
- future-proof for overlay rendering and spatial audio visualizations
- native on all tier-1 platforms
Platforms
| Platform | Status |
|---|---|
| Linux x86_64 | Phase 1 |
| Windows x86_64 | Phase 1 |
| macOS (Apple Silicon) | Phase 1 |
| macOS (Intel) | Phase 1 |
| Linux ARM64 | Phase 2 |
Audio pipeline
cpal (capture)
│ PCM f32 48000Hz
▼
RNNoise (noise suppression)
▼
AEC (echo cancellation)
▼
VAD (voice activity detection)
▼
opus encode
▼
LNEx UDP packet → voice-node
Playback:
LNEx UDP packet ← voice-node
▼
jitter buffer
▼
opus decode
▼
rodio (playback)
▼
cpal (output device)
Audio device selection is configurable in Settings → Voice and Settings → Audio Output.
Design system
The client follows the VNOX Hi-Fi Minimalism design system.
Full specification: docs/04-clients/design-system.md
Key decisions:
- monospace font throughout (IBM Plex Mono)
- warm dark palette (
#0d0d0dbase,#ff6b35accent) - no round server icons — square with abbreviation, 40px sidebar
- latency indicator bottom-left, next to identity
- no separate member list panel — voice users shown inline in channel list
Layout
┌──────────────────────────────────────────────────────┐
│ titlebar: dots · VNOX wordmark · connected node │
├────────┬──────────────────┬───────────────────────── │
│ nodes │ channels │ main (chat / voice) │
│ 40px │ 190px │ flex │
│ │ │ │
│ NC ◄ │ nightcore.lnex │ #general │
│ DV · │ ───────────── │ ───────────────────── │
│ GG │ # general ◄ │ messages │
│ VD │ # dev-talk │ │
│ + │ # plugins │ │
│ │ │ │
│ │ ~ lobby │ │
│ │ · raven │ ───────────────────── │
│ │ · 0xmist │ #general › [input] │
│ │ 🔇 lurker_7 │ │
│ │ ~ gaming │ │
│ │ ───────────── │ │
│ │ ● 12ms lnex v1 │ │
│ │ [YU] you 🎤 🎧 ⚙ │ │
└────────┴──────────────────┴──────────────────────────┘
Settings
Settings are stored locally. No settings are synced to the server.
Voice
- input device
- input volume
- noise suppression (RNNoise on/off)
- echo cancellation (on/off)
- activation mode: push-to-talk / voice activity / always on
- VAD threshold
- Opus bitrate (8–128k)
- Opus frame interval (10 / 20 / 40ms)
Audio output
- output device
- output volume
- jitter buffer size
- adaptive jitter buffer (on/off)
Network
- relay address
- auto relay selection
- UDP port
- force relay only (disable direct)
Identity
- view pubkey
- export keypair
- seed phrase backup
- rotate keypair
Appearance
- color scheme
- UI scale
- font
Keybinds
- push-to-talk key
- mute toggle
- deafen toggle
- overlay toggle
Plugins
- installed plugin list
- enable / disable per plugin
Advanced (debug)
- log level
- log voice packets
- show packet stats
- disable encryption (dev only)
Building from source
# Prerequisites: Rust stable, system audio libs
# Linux (Ubuntu/Debian)
apt install libasound2-dev libopus-dev
# macOS
brew install opus
# Build
git clone https://github.com/vnox/vnox
cd vnox/client
cargo build --release
# Binary
./target/release/vnox-client
Native client redesign (October 2026)
The current Slint shell uses neutral graphite surfaces, a server/channel sidebar, readable variable-height messages, a collapsible member panel, and persistent voice/profile controls below navigation. DM selection uses conversation IDs. Drafts are isolated per conversation and are cleared when leaving a server. Editable text, settings tabs, and keyfile forms survive UI synchronization.
Audio settings expose actual device lists, transmission mode, volume, and bitrate;
preferences are saved in the platform config directory at vnox/audio.json.
Push-to-talk has a hold control in the voice view. Encrypted identity vaults show
an unlock screen at startup. Message deletion asks for confirmation.
The UI has no demo members/messages. Threads, uploads, camera, screen sharing, and global Hub features remain future work. The existing client JSON transport still needs migration to the server's Protobuf/TLS contract. The server currently has a single channel membership per session, so independent server-side text and voice subscriptions are not guaranteed by the new local view state. Legacy E2EE DM sends are blocked with a retained draft until a verified nonce/message-ID contract is implemented; the client does not advertise unverified E2EE status.
For native headless rendering and binding regressions, run cargo test --workspace.
Set VNOX_UI_SNAPSHOT_DIR=/tmp/vnox-ui to emit test-only PPM screenshots for the
chat, settings, voice, and connection views. These screenshots use test fixtures
and do not connect to a server.
See Themes and profile photos for appearance presets, restricted CSS themes, and the local avatar cropper.