llama-cpp-desktop

Project Overview

Llama.cpp Desktop is a desktop control panel for local llama.cpp on Windows. Starting the service, configuring models, viewing logs, chatting directly, and connecting OpenAI-compatible clients are all in one window.

Key Features

• Direct local connection: Launches llama-server.exe directly from the original llama.cpp directory, with no strong dependency on extra launchers
• OpenAI compatible: Provides http://127.0.0.1:8080/v1 by default, accessible by OpenClaw, Claude Code, and other clients
• Desktop chat: Built-in web-style chat page with streaming replies, history, search, and message operations
• Attachment entry: Supports image, text, PDF, and other attachments; images can be previewed in chat
• Model info: Click the model tag to view the current model, context, GPU layers, and runtime parameters
• Terminal logs: View llama.cpp output in the client for troubleshooting startup and inference issues
• Tray background: Hides to the system tray when the window is closed; the service continues running in the background
• Parameter configuration: Supports model path, context, sampling, GPU layers, threads, and batch parameters

Target Audience

• Windows 10/11 users
• Developers who need to manage local llama.cpp inference services
• Users who prefer a graphical interface over the command line

Quick Start

1. Download Llama.cpp-Desktop.exe from Releases
2. Double-click to run
3. In settings, select the original llama.cpp directory or directly select llama-server.exe
4. Select your GGUF model file
5. Save the configuration and start the service; use built-in chat or connect an OpenAI-compatible client

Core Advantages

• Graphical interface, no need to memorize complex command-line parameters
• Built-in chat page, ready to use
• Tray background running without interfering with other work
• Can connect to external clients like OpenClaw and Claude Code

System Requirements

• OS: Windows 10 / 11
• Additional dependencies: Requires an existing usable llama.cpp Windows build directory and GGUF model files on the machine; the project does not bundle llama.cpp, model files, GPU drivers, or CUDA/Vulkan runtimes