Skip to main content
RCLI Waveform

​
What is RCLI?

RCLI (RunAnywhere Command Line Interface) is a complete STT + LLM + TTS pipeline running on Apple Silicon with Metal GPU. Control your Mac with voice or text, query local documents, and execute 43 macOS actions—all with sub-200ms end-to-end latency. No cloud, no API keys, 100% local inference.

Voice Pipeline

Complete STT, LLM, TTS pipeline with sub-200ms latency

macOS Actions

43 system actions controlled by voice or text

RAG System

Local document search with hybrid vector + BM25

Interactive TUI

Push-to-talk interface with live performance monitoring

​
Key Features

All AI processing runs on-device with Apple Silicon’s Metal GPU. No cloud API calls, no data leaves your Mac. Complete privacy and zero latency overhead from network requests.
From voice input to audio output in under 200ms on Apple Silicon. Optimized with system prompt KV caching, lock-free ring buffers, and sentence-level TTS scheduling.
Hot-swap between LLM models (Qwen3, LFM2), STT engines (Whisper, Parakeet), and TTS voices (Piper, Kokoro, KittenTTS) without restarting. All models run locally.
LLM-native tool calling with model-specific formats (Qwen3, LFM2). Execute 43 macOS actions including Spotify control, Safari automation, Notes creation, and system commands.
Index local documents (PDF, DOCX, TXT) and query them with ~4ms retrieval latency. Combines vector search (USearch HNSW) with BM25 full-text search using Reciprocal Rank Fusion.
Built with C++17 for Apple Silicon. Three-thread pipeline with lock-free ring buffers, 64MB pre-allocated memory pool, hardware profiling at startup, and token-budget conversation trimming.

​
Quick Start

Get RCLI running in 3 commands: 
brew tap RunanywhereAI/rcli https://github.com/RunanywhereAI/RCLI.git
brew install rcli
rcli setup
Then run rcli for interactive mode or rcli listen for continuous voice mode.

Installation Guide

Detailed installation instructions and system requirements

​
Example Commands

# Interactive TUI (push-to-talk + text)
rcli

# Continuous voice mode
rcli listen

# One-shot text commands
rcli ask "open Safari"
rcli ask "create a note called Meeting Notes"
rcli ask "play some jazz on Spotify"

# RAG document queries
rcli rag ingest ~/Documents/notes
rcli ask --rag ~/Library/RCLI/index "summarize the project plan"

# Model management
rcli models              # browse and switch models
rcli upgrade-llm         # upgrade to larger LLM
rcli voices              # manage TTS voices

# Benchmarking
rcli bench               # run all benchmarks
rcli bench --suite llm   # LLM only

​
System Requirements

macOS only — Apple Silicon (M1 or later), macOS 13+. RCLI uses Metal GPU for inference and CoreAudio for voice I/O.
  • Platform: macOS 13 Ventura or later
  • Chip: Apple Silicon (M1, M2, M3, or later)
  • Storage: ~1GB for default models (more for additional models)
  • RAM: 8GB minimum, 16GB recommended

​
What’s Next?

Quickstart

Get started with your first voice command in 5 minutes

Commands Overview

Learn all RCLI commands and their options

Architecture

Understand the voice pipeline and threading model

API Reference

Embed RCLI in your own applications with the C API

Build docs developers (and LLMs) love

Get started for freeTalk to us