Running large AI language models locally on a PC is a task that is becoming increasingly popular. To solve this problem, AMD has released Gaia, an open-source program designed to run local AI models on Windows computers.
Image source: amd.com
Gaia offers additional performance optimizations for machines with Ryzen AI processors, and its AI inference is powered by the Lemonade SDK. The software allows models to be tailored for a variety of purposes, from summarizing large texts to handling complex tasks that require reasoning. Gaia works through a Retrieval-Augmented Generation (RAG) component that combines AI with a knowledge base to provide more accurate, context-sensitive answers to queries.
RAG includes four agents:
- Simple Prompt Completion – provides direct interaction with the model, intended for testing and evaluation.
- Chaty — responsible for the chatbot interface.
- Clip – Searches YouTube with Q&A functionality.
- Joker – generates jokes and adds personality to the chatbot.
Gaia solves specialized AI problems using the Lemonade SDK, running in a variety of runtime environments. Lemonade exposes a web service for the language model and communicates with the Gaia app via an OpenAI-compatible REST API. The app acts as an AI agent that retrieves and processes data, “vectorizes” external content (such as data from GitHub, YouTube, and text files), and stores it in a local vector index. Simply put, Gaia improves user queries before they are processed by the AI model, improving the accuracy and relevance of responses.
A hybrid installer optimized for Ryzen processors, which allows Gaia to use the Ryzen AI’s built-in neural processing unit (NPU) and integrated graphics for computation. Running AI models locally has several advantages over cloud services: data remains secure, latency is reduced, and in some cases, performance is improved. Local models also do not require an internet connection.