RunAnywhere CLI

August 25, 2026 · View on GitHub

Run open models on your machine.

rcli pull qwen3
rcli run qwen3

Chat, vision, speech, and embeddings — all local. Nothing leaves the device.

Install

macOS (Apple Silicon)

brew install runanywhereai/rcli/rcli

or

curl -fsSL https://raw.githubusercontent.com/RunanywhereAI/RCLI/main/install.sh | sh

Windows (x64)

irm https://raw.githubusercontent.com/RunanywhereAI/RCLI/main/install.ps1 | iex

Linux (x86_64)

curl -fsSL https://raw.githubusercontent.com/RunanywhereAI/RCLI/main/install.sh | sh

Get started

rcli pull qwen3          # download
rcli run qwen3           # chat
rcli run qwen3 "Hello"   # one-shot
rcli serve qwen3         # OpenAI-compatible API on :8080 (macOS/Linux)

rcli models list --all is the full catalog. Short names work everywhere (qwen3, llama3.2, whisper-tiny, piper, …). Any Hugging Face GGUF works too:

rcli pull hf.co/Qwen/Qwen3-0.6B-GGUF/Qwen3-0.6B-Q8_0.gguf

Backends

One rcli binary. The kit picks the engine; you do not.

BackendmacOS (Apple Silicon)Windows x64Windows ARM64Linux x64
llama.cppyesyesyes
MLXyes
Sherpa-ONNXyesyesyes
ONNX Runtimeyesyesyes
NeuRT (Apple Neural Engine + Core ML)overlay
QHexRT (Qualcomm Hexagon NPU)overlay

rcli backends prints what this binary actually registered.

MLX is Apple GPU. NeuRT is Apple Neural Engine — Core ML is the stack NeuRT uses, not a separate backend. Image generation (sd15) runs on NeuRT. NeuRT and QHexRT are private overlays, not in the public bottle. QHexRT is Snapdragon NPU on Windows ARM64 only (x64 Windows has no Hexagon path). llama.cpp / Sherpa / ONNX do not configure on MSVC ARM64 today, so the public Windows ARM64 kit is commons + the desktop adapter.

Models

Catalog models are grouped by the org that trains them. GGUF rows run on llama.cpp (macOS, Windows x64, Linux). mlx-* rows run on Apple Silicon only.

Language

OrgFamiliesTry
Alibaba QwenQwen3, Qwen3.6, Qwen3.8qwen3, mlx-qwen3
MetaLlama 3.2llama3.2, mlx-llama3.2
GoogleGemma 4gemma4-e2b, mlx-gemma4-e2b
Hugging FaceSmolLM2smollm2
Liquid AILFM2lfm2
IBMGranite 4.1granite4.1-3b, mlx-granite4.1-3b
NVIDIANemotronmlx-nemotron-nano
PrismMLBonsai, Ternary-Bonsaibonsai-1.7b, mlx-bonsai-1.7b
DeepGroveMaple Previewmaple-preview, mlx-maple-preview

Vision

OrgFamiliesTry
Hugging FaceSmolVLM2smolvlm2
Alibaba QwenQwen2-VLqwen2-vl, mlx-qwen2-vl
Liquid AILFM2-VL, LFM2.5-VLlfm2-vl, mlx-lfm2.5-vl
AppleFastVLMmlx-fastvlm
MicrosoftFara 1.5 (computer use)fara
MetaMuse Glimmermuse-glimmer
NVIDIANemotron Omninemotron-omni
rcli vlm generate --model smolvlm2 --image photo.png "What is in this picture?"

Speech

OrgFamiliesRoleTry
OpenAIWhisperSTTwhisper-tiny
NVIDIAParakeet, Canary, Nemotron ASRSTTparakeet-tdt-v2
Alibaba QwenQwen3-ASR / Qwen3-TTSSTT / TTS (MLX)mlx-qwen3-asr
rhasspyPiperTTSpiper
SupertoneSupertonicTTSsupertonic
ZhipuGLM-ASRSTT (MLX)mlx-glm-asr
SileroSileroVADsilero
rcli tts synthesize "Hello from the device." -o hello.wav
rcli stt transcribe hello.wav

Embeddings, rerank, other

OrgFamiliesRoleTry
NVIDIANemotron Embed, Llama-Nemotron Embedembeddingsnemotron-3-embed
Alibaba QwenQwen3 Embeddingembeddings (MLX)mlx-qwen3-embed
sentence-transformersMiniLMembeddingsminilm
BAAIBGE Rerankerrerankbge-reranker
NVIDIASortformerdiarizationsortformer
NVIDIA / Hugging FaceSegFormersegmentationsegformer
Stability AI / AppleStable Diffusion 1.5image gen (NeuRT)sd15

macOS vs Windows

macOS Apple Silicon is the full product: llama.cpp + MLX + Sherpa + ONNX in one binary. Pull qwen3 (CPU/Metal GGUF) or mlx-qwen3 (Apple GPU). Image generation (sd15) is NeuRT (Apple Neural Engine / Core ML). NeuRT links when the private overlay is present.

Windows x64 runs the GGUF / ONNX / Sherpa catalog: Qwen, Llama, Gemma, Granite, Whisper, Piper, MiniLM, and the rest of the non-mlx-* rows. No MLX, no NeuRT, no QHexRT.

Windows ARM64 (Snapdragon): public kit has no llama.cpp/ONNX/Sherpa yet. With the QHexRT overlay, Hexagon NPU models run on device. Do not expect mlx-* or sd15 here.

rcli serve is macOS and Linux.

Commands

rcli run / rcli chatchat (REPL with no prompt)
rcli pull / rcli models downloaddownload
rcli list / rcli lslocal models (--all = catalog)
rcli showone model
rcli rmdelete
rcli llm generate / streamcompletion
rcli vlm generate --imagevision
rcli stt transcribespeech → text
rcli tts synthesizetext → WAV
rcli vad detectvoice activity
rcli embedembeddings
rcli rerankrerank documents
rcli image generatetext → image (NeuRT / Apple Silicon)
rcli serveOpenAI-compatible HTTP (macOS/Linux)
rcli backendsregistered engines
rcli infoversions and paths

rcli --help and rcli <command> --help cover the rest.

Build from source

Stage a C++ desktop kit from runanywhere-sdks. The pin is cmake/sdk-pin.cmake (RCLI_PINNED_SDK_VERSION).

C++-only (rcli-cxx on Apple; rcli elsewhere):

cmake -B build -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_PREFIX_PATH=/path/to/kit
cmake --build build
./build/rcli version   # ./build/rcli-cxx on Apple
./build/rcli backends

Apple Silicon product binary is the Swift MLX host (build/rcli). Independent clones need the SDK Swift tree (RCLI_SDK_SWIFT_PATH) and RCLI_APPLE_MLX_HOST=ON (the default):

export RCLI_SDK_SWIFT_PATH=/path/to/runanywhere-sdks
cmake -B build -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_PREFIX_PATH=/path/to/kit
cmake --build build
# or: scripts/build-mlx.sh build
./build/rcli version
./build/rcli backends

See CONTRIBUTING.md.

Docs

MIT. See LICENSE.