AMD released GAIA 0.25 on Sunday, Oct. 18, according to the project’s GitHub release page. The open-source project for building and running local AI agents on Ryzen AI hardware started out as “Generative AI Is Awesome” and now goes by the tag line “AMD’s open-source agent factory,” Phoronix noted in its Oct. 19 report.
The biggest practical change is size. PyTorch, transformers and accelerate are gone, and the desktop app’s Python environment on Windows drops from about 0.9GB to about 0.1GB. Windows users also get a single installer for the desktop app and the terminal, with a GAIA Windows Terminal profile.
A fix for 16GB machines
The release also fixes a first-launch crash on PCs with 16GB of RAM. According to the pull request behind the fix, the desktop app crashed on a 16GB Linux laptop when the kernel killed it for running out of memory, and only a second launch worked. GAIA now checks free memory before it loads a model and shrinks the context window to fit, down to no less than 32K. If even that won’t fit, it says how much memory it needs and what to close.
Agent and model changes
Running gaia chat now starts the flagship GAIA agent instead of the retired ChatAgent, and gaia talk transcribes and speaks through AMD’s Lemonade local AI server, using Whisper and kokoro-v1. Setup installs GAIA agent 0.3.0.
A new chat caches its prompt as it opens, which AMD says cuts the time to answer a first “hello” on Qwen3.6 from about 43 seconds to about 21. Long Agent UI turns keep running as long as they make progress and stop only after 75 minutes without any, a limit users can change in seconds with the GAIA_STREAM_TIMEOUT variable.
On the model side, gaia init now picks Qwen3.6-35B-A3B on a Strix Halo system with 64GB or more, or on a 32GB GPU, and keeps Gemma-4-E4B elsewhere. Qwen3.8 Flash Next is an opt-in on 128GB Strix Halo machines, and DeepSeek V4.1 Flash is the new default cloud model on the AMD LLM Gateway and Fireworks.