Ava PLS

Ava is an open-source desktop application for running language models locally on your computer. It provides a batteries-included graphical user interface (GUI) built around llama.cpp, enabling users to run language models offline with a convenient local workflow. The project uses Zig and C++ (llama.cpp), with SQLite for data handling and a modern frontend stack (Preact, Preact Signals, Twind) for a responsive UI. Ava can be downloaded as prebuilt artifacts or built from source, including a headless mode option for non-GUI usage.

How Ava Works

• Local model execution: Run language models directly on your machine without requiring cloud access.
• GUI-based workflow: A user-friendly desktop interface to configure and execute models, manage sessions, and view outputs.
• Build options: Download prebuilt binaries or build from source using Zig; supports standard build commands.
• Headless mode: Use -Dheadless=true to run Ava without the GUI for automation or server-style tasks.

Getting Started

1. Download or build Ava. Obtain prebuilt artifacts from GitHub Actions or build locally with Zig.
2. Launch Ava. Open the GUI to start configuring models, selecting weights, and issuing prompts.
3. Run models locally. Execute inference tasks directly on your computer, with results rendered in the app.
4. Optional headless run. Build with -Dheadless=true to operate Ava in headless mode for scripting and automation.

Tech Stack

• Zig programming language
• C++ (llama.cpp) for model execution
• SQLite for data management
• Frontend: Preact, Preact Signals, Twind

Safety and Licensing

• Ava is MIT-licensed and intended for local, offline use of language models. Respect model licenses and data privacy when handling prompts and outputs.

Core Features

• Local execution of language models with a GUI
• Open-source MIT-licensed project
• Built around llama.cpp for efficient on-device inference
• Zig-based build system with support for zig build run and zig build run -Dheadless=true
• Integrated frontend stack (Preact, Preact Signals, Twind) for a responsive UI
• SQLite for lightweight local data management
• Prebuilt artifacts or source builds from GitHub Actions
• Headless mode for automation and scripting