Native AI assistant for Linux with its own window, system tray and global hotkey. No terminal dependency.
- Streaming chat — Real-time responses with Markdown rendering, syntax highlighting and copy-to-clipboard code blocks
- Multi-provider — GitHub Copilot, OpenAI, Venice, Groq, OpenRouter, Gemini, xAI, Custom (Ollama/LM Studio)
- OpenClaw skills catalog — Search and import compatible community skills directly from Settings
- Agent mode — Execute actions on your PC: create/edit/delete files, open apps, web search, organize folders and generate images
- Voice input and TTS — Dictate prompts by microphone and read assistant responses aloud
- Drag & drop files — Drop PDF, DOCX, TXT, images directly into the chat
- Conversation history — Persisted locally with export to Markdown
- Dark mode — Light/dark theme toggle
- System tray — Minimizes to the tray with context menu
- Global hotkey —
Super+Shift+Bto show/hide the window from anywhere - Autostart — Option to launch on system startup (XDG autostart)
- Web search — Searches DuckDuckGo and brings results inside the chat
- Multi-language — 11 languages: English, Português, Español, Русский, 日本語, 中文, العربية, Deutsch, Français, Italiano, हिन्दी
- Account rotation — Multiple API keys with automatic rotation on rate limit
| Layer | Technology |
|---|---|
| Window | tao + wry (WebKitGTK) |
| System tray | tray-icon |
| Global hotkey | global-hotkey |
| Core | Rust (business logic, providers, agent, streaming) |
| Frontend | React 18 + TypeScript + Vite + Tailwind CSS v4 |
| IPC | JSON postMessage via custom brother://app/ protocol |
- Linux (Ubuntu 22.04+ recommended)
- Node.js 18+
- Rust 1.75+
- System dependencies:
sudo apt install libwebkit2gtk-4.1-dev libgtk-3-dev libayatana-appindicator3-dev
- Optional TTS engines for read-aloud:
sudo apt install espeak-ng speech-dispatcher
# Install frontend dependencies
npm install
# Build the frontend
npm run build
# Run in dev mode
cargo run -p brother-shell# Optimized build + .deb package
./scripts/build-release.sh
# Or manually:
npm run build
cargo build --release -p brother-shellbrother-core/ # Business logic (providers, agent, config, streaming)
brother-shell/ # Native Linux shell (tao + wry + tray + hotkey)
linux-cli/ # Alternative CLI (terminal)
skills/ # Local skills loaded in development and portable installs
src/ # React frontend (components, styles, types)
scripts/ # Build scripts
- Brother now has a local skills base for prompt augmentation and task specialization.
- Brother can also search and import compatible community skills from OpenClaw, making external skill catalogs available inside Settings.
- Skills are loaded from
skills/in the current project, from a siblingskills/directory near the executable, or from~/.config/copilot-assistente/skills/. - Each skill lives in its own folder and must provide a
SKILL.mdwithname,descriptionand Markdown instructions. - Supported manifest fields now include
metadatawith a Brother JSON object containingversion,source,tools,permissions,installRequired,requiresApprovalandautoActivate. - Skills that require permissions or installation are cataloged but not auto-activated; the runtime flags them as approval-only.
- Settings now include a skills browser with local installed skills, OpenClaw catalog search, and one-click install for compatible OpenClaw skills when supported by the native host.
- In browser preview, the UI falls back to a built-in catalog reader so skills search still works even before restarting the native shell.
- Agent responses now include the executed action label, such as
open_application,open_browser_search, orweb_search, to make behavior easier to inspect.
- Zero telemetry — No data is sent to project servers
- Local config — API keys and history stay in
~/.config/copilot-assistente/ - Only external communication — The AI provider API you chose
- 100% offline option — Use Custom provider pointing to a local Ollama instance
LGPL