AMD's GAIA Finally Has A Nice Multi-Device Experience For AI
AMD's GAIA open-source project geared for building AI agents that run locally on your PC is out with a significant new feature release for Windows and Linux systems.
With the now-released AMD GAIA 0.20, AI agents have a choice of hardware with better handling for effectively leveraging your AMD CPU, GPU, and/or NPU. Up to this point AMD's GAIA would default to AI inferencing on the supported GPU if available with a capable Llama.cpp back-end. GAIA lacked a way to choose an alternate device if preferring to use your low-power Ryzen AI NPU or CPU for other reasons. With GAIA 0.20, there is now support for per-agent device selection. Each AI agent can declare which accelerator devices it supports via new "DeviceConfig" tuples. Users are also able to select a device per-agent via their Agent UI or the new "--device" command-line flag.
The Llama.cpp GPU back-end will still be the default but now there is more opportunities for better heuristics in modern multi-device systems.
GAIA 0.20 also has greater controls over what AI agents can see or do on the system. There is also a Agent Hub TUI for a text-based, terminal-native hub for browsing, searching, and managing agents. There is also Microsoft PowerPoint PPTX RAG support, security and first-boot robustness improvements, and up to 10x faster email organization.
For downloads or to learn more about AMD GAIA 0.20, see GitHub.
With the now-released AMD GAIA 0.20, AI agents have a choice of hardware with better handling for effectively leveraging your AMD CPU, GPU, and/or NPU. Up to this point AMD's GAIA would default to AI inferencing on the supported GPU if available with a capable Llama.cpp back-end. GAIA lacked a way to choose an alternate device if preferring to use your low-power Ryzen AI NPU or CPU for other reasons. With GAIA 0.20, there is now support for per-agent device selection. Each AI agent can declare which accelerator devices it supports via new "DeviceConfig" tuples. Users are also able to select a device per-agent via their Agent UI or the new "--device" command-line flag.
The Llama.cpp GPU back-end will still be the default but now there is more opportunities for better heuristics in modern multi-device systems.
GAIA 0.20 also has greater controls over what AI agents can see or do on the system. There is also a Agent Hub TUI for a text-based, terminal-native hub for browsing, searching, and managing agents. There is also Microsoft PowerPoint PPTX RAG support, security and first-boot robustness improvements, and up to 10x faster email organization.
For downloads or to learn more about AMD GAIA 0.20, see GitHub.