Pre-built binary releases of the MetalRT GPU inference engine for macOS on Apple Silicon.
MetalRT is a closed-source, high-performance inference engine optimized for Metal GPU. It powers the GPU acceleration features in RCLI.
MetalRT is automatically downloaded and installed by RCLI:
rcli setup # choose MetalRT or Both engines
rcli metalrt install # install/update MetalRT separatelyOr download directly from Releases.
Each release tarball contains:
| File | Description |
|---|---|
libmetalrt.dylib |
MetalRT shared library (LLM + STT + TTS inference) |
default.metallib |
Compiled Metal GPU kernels |
metalrt_c_api.h |
C API header for consumers |
checksums.sha256 |
SHA-256 checksums for verification |
cd ~/Library/RCLI/engines
shasum -a 256 -c checksums.sha256
codesign -v libmetalrt.dylib- macOS 14+ (Sonoma or later)
- Apple Silicon (M1, M2, M3, M4)
- ~10 MB disk space for engine binaries
Current release: v0.2.1 (ABI version 2)
See LATEST_VERSION file for the version string used by rcli metalrt install.
Proprietary. See LICENSE for terms.
MetalRT is developed by RunAnywhere, Inc. The binary may be used with RCLI and compatible applications. Redistribution and reverse engineering are prohibited.
- Licensing & partnerships: founder@runanywhere.ai
- RCLI issues: https://github.com/RunanywhereAI/RCLI/issues
- Website: https://runanywhere.ai
