Version 3.0.0 · Tauri 2 + React 18 + Vite
Native desktop shell for the KERNEL cognitive operating environment.
A dark, high-density terminal-inspired interface that unifies multi-agent orchestration, policy enforcement, chat, research, data analysis, and hardware workflows under a single window.
KERNEL Desktop is a Tauri 2 application that packages a large single-file React UI (src/KernelDesktop.jsx) together with a minimal Rust backend.
It is designed to run both as a native desktop app and as a browser fallback when Tauri APIs are unavailable.
| Domain | Description |
|---|---|
| Chat / Multi-agent | Route tasks across specialized KERNEL roles (DATA, RESEARCH, HW, etc.) |
| Policy engine | Visual policy cards, authority levels, and constraint guards |
| Agent sidebar | Live agent list, history, eco chips, new-chat controls |
| System bridge | Tauri invoke + plugins (HTTP, FS, Dialog, Store, Shell) with graceful browser fallback |
| Config persistence | Tauri Store (kernel-config.json) or localStorage fallback |
| Threat / chaos testing | Built-in vectors (e.g. CL-1 Authority & Trust Subversion) and scenario cycles |
The visual language is deliberately monospace / jet-black / cyan (#00ffd5) — a “command deck” aesthetic.
kernel-desktop/
├── index.html # Minimal shell, #root mount
├── package.json # React 18 + Tauri 2 plugins
├── vite.config.js # Vite + React plugin, port 5173
├── src/
│ ├── main.jsx # ReactDOM entry → <KernelDesktop />
│ └── KernelDesktop.jsx # Entire UI + Tauri bridge (≈189 kB)
└── src-tauri/
├── Cargo.toml # Tauri 2 + plugins
├── tauri.conf.json # Window 1400×900, CSP, plugin scopes
├── capabilities/default.json
├── build.rs
└── src/
├── lib.rs # get_system_info command + plugin wiring
└── main.rs
KernelDesktop.jsx contains a self-contained bridge that:
- Detects
window.__TAURI_INTERNALS__ - Dynamically imports
@tauri-apps/api/core+ plugins - Falls back to browser
fetch/localStoragewhen running outside Tauri
This makes the same artifact usable for both desktop packaging and pure web preview.
| Tool | Version / Notes |
|---|---|
| Node.js | ≥ 18 (tested on 24) |
| npm | ≥ 9 |
| Rust | ≥ 1.77 recommended (1.75 may fail on some Tauri 2 transitive crates requiring edition2024) |
| Tauri CLI | npm install -g @tauri-apps/cli or use local npx tauri |
| System deps | See Tauri prerequisites for your OS |
# 1. Install frontend dependencies
npm install
# 2. Dev mode (Vite only – browser)
npm run dev
# → http://localhost:5173
# 3. Full desktop (Tauri)
npm run tauri devnpm run tauri buildArtifacts appear under src-tauri/target/release/bundle/.
-
Export name mismatch
src/KernelDesktop.jsxcurrently exportsKernelV3, whilesrc/main.jsximportsKernelDesktop.
Fix: rename the default export (or the import) so they match. -
Rust toolchain
Cargo check on Rust 1.75 can fail because some dependency crates now requireedition2024.
Upgrade to a recent stable Rust (≥ 1.80 / 1.81) if you hit this. -
Icons
tauri.conf.jsonreferences icon files that are not yet present in the repository.
Addsrc-tauri/icons/(32×32, 128×128, etc.) before shipping a release bundle. -
Large single-file component
All UI logic lives in one 189 kB JSX file. Future work may split it into proper modules for maintainability.
- Window: 1400 × 900, min 900 × 600, centered, resizable
- CSP: Allows self + HTTPS + localhost / private LAN
- FS scope:
$APPDATA/**and$HOME/.kernel/** - HTTP scope: HTTPS anywhere + localhost + common private ranges
- Store: Persistent key-value (
kernel-config.json) - Shell:
openpermission enabled
See src-tauri/tauri.conf.json and src-tauri/capabilities/default.json for the full capability set.
- The UI is intentionally dense and keyboard-oriented.
- All styling is inline (no CSS modules / Tailwind) for maximum portability of the single JSX artifact.
- When running under Tauri, the bridge opens DevTools automatically in debug builds.
- System info is exposed via the Rust command
get_system_info.
© 2026 Benjamin Hubbard / Terracare Inc.
All rights reserved. Not licensed for redistribution unless explicitly stated otherwise.
◈ KERNEL — command deck for multi-agent cognition.