commit 42f033ebaa16d43a4177a7ecad46c9a2ee1ae29d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 14:51:38 2026 +0200

sci-ml/lemonade: add 10.4.0

commit 3c36ed055d0afc2bd7e490e1b1102171e397f8b8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 21:47:06 2026 +0200

sci-ml/lemonade: new package, add 9999 + 10.3.0

Local AI server: optimized LLM inference on AMD NPU + GPU. C++ server
core (lemond + lemonade CLI + the deprecated lemonade-server shim)
installed under /usr to match lemond's get_resource_path() search
list — /opt prefix doesn't fit because the resolver expects
/opt/share/lemonade-server/, not /opt/lemonade/share/.

Tauri desktop wrapper and the bundled web frontend are disabled at
configure time (BUILD_TAURI_APP=OFF + BUILD_WEB_APP=OFF) for a
server-only build; both can move behind USE flags later.

PROPERTIES=live + RESTRICT=network-sandbox because cpp-httplib and
libwebsockets fall back to FetchContent under default ::gentoo
system deps (cpp-httplib lacks a .pc file; libwebsockets isn't
installed by default in this overlay). The FetchContent-built
libwebsockets.a is dropped from the install image — lemond linked
it statically already, no consumers.

USE=fastflowlm pulls in sci-ml/fastflowlm as the NPU runtime
backend; verified end-to-end (lemond + 'lemonade backends' lists
flm npu installed v0.9.39).