v0.1.71

Fixed

• NVIDIA driver 610+ silently landed the managed backend on the wrong GPU. Driver 610 renamed the nvidia-smi header fields, the CUDA version parser stopped matching, and the backend selector fell back to the Vulkan build, which re-admits integrated GPUs into auto-selection. The parser now accepts both header generations (#55 by @sachin-detrax, fixes #54).

If your Windows install already picked the wrong backend: this fix cannot repair an existing wrong download (every backend zip ships the same llama-server.exe, so the installer sees it as already present). Delete the backend directory and run atomic-agent models update to force a clean re-download.

Updating

macOS / Linux

curl -fsSL https://atomicagent.io/install | sh

Windows (PowerShell)

irm https://atomicagent.io/install.ps1 | iex

Existing installs update in place; the TUI prompts for a restart afterwards.

Thanks @sachin-detrax for the report, the diagnosis and the fix.

Full Changelog: v0.1.69...v0.1.71