runanywhere-sdks

Desktop kits for 0.20.39. D1 and GLiNER text decisions.

Fixed C++ desktop kits: CPU-only context refit honored (no 4096 clamp), no AVX-512/AMX codegen, no SVE; restores llama.cpp to the windows-arm64 kit.

Windows ARM64 kit now ships the llamacpp backend (rac_backend_llamacpp.lib), built from feat/windows-arm64-llamacpp at 640021724 (ggml MSVC-ARM fix via llama.cpp fork tag runanywhere-b10453.7). macOS

Desktop kits rebuilt with the glibc-2.35-safe Sherpa-ONNX desktop build (v1.13.5-rac-desktop.6). The -2 linux-x64 kit bundled a sherpa built on ubuntu-24.04 that required GLIBC_2.38 / GLIBCXX_3.4.32;

v0.20.38 — decision support (Apple binaries + desktop kits) Apple XCFrameworks (this release) Nine verified archives, built from kit/0.20.38-decision (e595580a7): - RACommons — ABI v13 (rac_decis

Desktop C++ kits for RunAnywhere 0.20.38 with the local decision primitive, consumed by the wally CLI (RCLI) via find_package(RunAnywhere). Same five kits as cpp-desktop-v0.20.38-1 (bzip2 bundled, Ub

Same kits as cpp-desktop-v0.20.38. Only linux-arm64 changed: its two Sherpa libraries are rebuilt on Ubuntu 22.04, so the arm64 build loads on glibc 2.35 (Ubuntu 22.04, Debian 12) instead of needing g

C++ desktop kits for Wally local coding harnesses. - Reports the context actually loaded after device fitting. - Reports MLX context from the loaded model configuration. - Includes macOS ARM64, Linux

What's Changed fix(web): restore surviving capability owners after backend teardown by @shubhamsinnh in https://github.com/RunanywhereAI/runanywhere-sdks/pull/714 fix(commons): propagate diffusion

Desktop C++ kits for RunAnywhere 0.20.37, consumed by the wally CLI (RCLI) via find_package(RunAnywhere). Ships the three kits RCLI pins: macos-arm64, windows-x64, windows-arm64. This is a desktop-ki