Nouveautés

CortexMind 4.0.0

GPU acceleration everywhere. Memory search now runs on the GPU on every platform: Metal on Apple Silicon, and any Vulkan-capable card on Linux and Windows, including NVIDIA, AMD and Intel Arc. Nothing extra to download: the inference engine ships inside the app.

Picks the right device on its own. On first launch CortexMind times your CPU against every GPU it can see and keeps the fastest. A GPU that would be slower than your CPU is never used. memory_status shows exactly which device each model is running on.

Lighter on your machine. About 2 GB of GPU memory when a GPU is in use, the reranker runs at full precision, and the first launch after upgrading removes the old model files, freeing several gigabytes of disk.

Faster on Apple Silicon. Memory search on an M-series Mac is roughly twice as fast as in 3.8.

Full changelog: https://cortexmind.net/changelog

Signed updater artifacts are attached. Existing installs update automatically.