Nous Research Adds One-Click Local Model Setup to Hermes Desktop
2026-09-06 03:12Products & Apps🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
3mentions
SummaryAI generated
On September 5, 2026, MarkTechPost reported that Nous Research introduced the “one-click local model configuration” feature in Hermes Desktop. This feature supports macOS 12+., Windows 10/11, and Linux systems. It automatically reads hardware information, selects appropriate models, downloads weights, and configures inference runtime (covering CUDA, Metal, Vulkan, HIP, and CPU) without the need for user login. Hermes manages the official llama.cpp build version and conducts memory adaptation assessments based on the user’s machine before downloading, indicating the status with colors. If the model exceeds the available memory, the system will unload the excess portion to system RAM according to the principle of priority for expert weights and lastly for attention cache. After 15 minutes of idleness, it will automatically clean up.
Nous Research has added a one-click local model configuration function in Hermes Desktop. This function automatically reads the user’s hardware, selects suitable models, downloads weights, and configures inference operations, supporting macOS 12+, Windows 10/11, and Linux systems without the need for an account. Hermes manages the official llama.cpp build versions, with backends covering CUDA, Metal, Vulkan, HIP, and CPU. Each model undergoes memory adaptation evaluation based on the user’s machine before downloading, with status indicated by green (full GPU), amber (overflowing system RAM), or red (unable to run). It is ensured that recommended models have at least 64K context windows. When a model exceeds the memory limit, Hermes will unload the excess portion into system memory according to the principle of expert weights first and attention cache last, and automatically uninstall the model after 15 minutes of idle time.