oClapp

A cross-platform desktop app for local LLM inference, powered by llama.cpp.

Features

• Model Browser: Search and download GGUF models from HuggingFace
• Local Inference: Run an OpenAI-compatible API server via llama.cpp
• Coding Agent Integration: Launch Claude Code, Codex CLI, and other agents with oClapp launch
• Settings UI: Curated and advanced parameter tuning with persistence
• Cross-Platform: Native desktop app for macOS, Linux, and Windows

Architecture

Cargo Workspace
├── oclapp-core    # Shared logic: models, server lifecycle, settings
├── oclapp-cli     # CLI binary: `oClapp launch`, `oClapp serve`
└── oclapp-tauri   # Desktop app: Tauri v2 + React + TypeScript

Development

Prerequisites

• Rust (latest stable)
• Node.js 20+
• Platform-specific dependencies for Tauri:
  • macOS: Xcode Command Line Tools
  • Linux: libgtk-3-dev, libwebkit2gtk-4.1-dev, libappindicator3-dev, librsvg2-dev
  • Windows: Visual Studio Build Tools with C++ workload

Setup

# Clone the repository
git clone https://github.com/your-org/oclapp.git
cd oclapp

# Install frontend dependencies
cd frontend && npm install && cd ..

# Build the entire workspace
cargo build

Running the Desktop App

# Development mode (hot reload)
cargo tauri dev

# Production build
cargo tauri build

Running the CLI

# List local models
cargo run --bin oclapp -- models

# Start the inference server
cargo run --bin oclapp -- serve --model llama-2-7b

# Launch Claude Code with a model
cargo run --bin oclapp -- launch claude --model llama-2-7b

Distribution

Platform-specific build scripts are provided in scripts/:

# macOS
./scripts/build-macos.sh

# Linux
./scripts/build-linux.sh

# Windows
.\scripts\build-windows.ps1

CI builds are configured via GitHub Actions (.github/workflows/build.yml) and triggered on pushes to main and version tags.

License

MIT