March 22 - Thursday, local time.AMD Announcing the launch of applications designed to run locally on the Large Language Model (LLM) GAIA(pronounced /ˈɡaɪ.ə/) and is open-sourced on Github, and is currently limited to the Windows platform.
GAIA is a generative AI application that enables localized private running of LLMs on Windows PCs and is optimized for the Rui Long AI 300 series processors, marking another important placement of AMD in the edge computing space.
Running GAIA on the NPU improves performance on AI-specific tasks. Starting with Ryzen AI software version 1.3, there is hybrid support for deploying quantized LLMs that use both the NPU and core graphics.
Its core components are developed based on ONNX TurnkeyML's Lemonade SDK, and it adopts the Retrieval Augmented Generation (RAG) technology framework, which realizes real-time retrieval and analysis of external data such as YouTube, GitHub, and so on through local vector indexes.
The system workflow contains three modules: LLM Connector, LlamaIndex RAG Pipeline, and Agent Web Server, which support mainstream models such as Llama and Phi. The function module contains four major AI intelligences:
- Chaty: a chatbot that supports historical conversations
- Clip: YouTube Video Search & Q&A Experts
- Joker: a joke generator with a humor style
- Simple Prompt: Model Direct Interaction Testing Tool
On the hardware side, AMD offers two installation options: Normal and Hybrid. The regular installer is compatible with any Windows PC (including non-AMD hardware devices), and cross-platform support is realized through the Ollama backend.
Hybrid installers are optimized for devices with the Raidon AI 300 series processors and intelligently distribute the computational load between the NPU and integrated graphics.AMD said that the Ryzen AI software has supported hybrid deployments of quantitative models between the NPU and integrated graphics since version 1.3.
Privacy and Performance Advantage Aspects:
- Localized processing ensures that data from sensitive industries such as healthcare and finance does not leave the country
- Reduced response latency to milliseconds (40-60% compared to cloud-based solutions)
- NPU-specific architecture delivers performance-per-watt gains and supports offline environments
The project uses the MIT open source protocol, and the codebase is open on GitHub. AMD specifically encourages developers to contribute custom agent modules, and several industry custom agents are currently in development.
Additionally, while GAIA does not currently support Linux / macOS, official AMD documentation reveals that multi-platform support may be expanded in the future.
1AI with Github address:
https://github.com/amd/gaia
