AMD GAIA 0.25 is out this week as the latest feature release for this open-source project that began as "Generative AI Is Awesome" and is now going on a tag line of "AMD's open-source agent factory" in building/using local AI agents on AMD Ryzen AI hardware.
The GAIA 0.25 brings a number of usability enhancements and other changes for improving the user experience. Running gaia chat is now defaulting to the flagship GAIA agent rather than their retired ChatAgent. The gaia talk sub-command is now transcribing and speaking via leveraging AMD's Lemonade local AI server project with Whisper and kokoro-v1 on the back-end.
GAIA 0.25 also now has a much smaller install with dropping unnecessary dependencies that the Python environment went from ~0.9GB down to ~0.1GB on Windows.
AMD GAIA 0.25 also addresses a long-standing obstacle where trying to run the software on a system with 16GB of RAM would run out of memory on the app's first launch. But on subsequent launches it would work. GAIA is now checking available free RAM before loading a model in order to shrink the context window to fit. This stems from a merge request recently that was motivated over the app's first launch crashing on Linux laptops with 16GB of RAM and being killed by the kernel due out-of-memory.
GAIA 0.25 also brings numerous agent and model improvements, Windows install/execution improvements, various optimizations for AMD Strix Halo, and more.
The lengthy list of GAIA 0.25 changes can be found via GitHub.