[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-75150":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":25,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":44,"readmeContent":45,"aiSummary":46,"trendingCount":16,"starSnapshotCount":16,"syncStatus":18,"lastSyncTime":47,"discoverSource":48},75150,"RCLI","RunanywhereAI\u002FRCLI","RunanywhereAI","Talk to your Mac, query your docs, no cloud required. On-device voice AI + RAG","https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002Frunanywhere-sdks",null,"C++",1523,83,6,10,0,1,2,17,3,58.47,"MIT License",false,"main",true,[27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"ai-assistant","apple-silicon","kitten-tts","kokoro-tts","lfm2","llama-cpp","llm","local-ai","metal","on-device-ai","parakeet","qwen3","rag","speech-to-text","text-to-speech","tool-calling","voice-assistant","2026-06-12 04:01:18","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Frcli_waveform.gif\" alt=\"RCLI Waveform\" width=\"700\" \u002F>\n  \u003Cbr>\n  \u003Cstrong>Talk to your Mac, query your docs, no cloud required.\u003C\u002Fstrong>\n  \u003Cbr>\u003Cbr>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fplatform-macOS-blue\" alt=\"macOS\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fchip-Apple_Silicon-black\" alt=\"Apple Silicon\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Finference-100%25_local-green\" alt=\"Local\">\u003C\u002Fa>\n  \u003Ca href=\"LICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-blue\" alt=\"MIT\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n**RCLI** is an on-device voice AI for macOS. A complete STT + LLM + TTS + VLM pipeline running natively on Apple Silicon — 40 macOS actions via voice, local RAG over your documents, on-device vision (camera & screen analysis), sub-200ms end-to-end latency. No cloud, no API keys.\n\nPowered by [MetalRT](#metalrt-gpu-engine), a proprietary GPU inference engine built by [RunAnywhere, Inc.](https:\u002F\u002Frunanywhere.ai) specifically for Apple Silicon.\n\n## Demo\n\n> Real-time screen recordings on Apple Silicon — no cloud, no edits, no tricks.\n\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\" align=\"center\">\n\u003Cstrong>Voice Conversation\u003C\u002Fstrong>\u003Cbr>\n\u003Cem>Talk naturally — RCLI listens, understands, and responds on-device.\u003C\u002Fem>\u003Cbr>\u003Cbr>\n\u003Ca href=\"https:\u002F\u002Fyoutu.be\u002FqeardCENcV0\">\n\u003Cimg src=\"assets\u002Fdemos\u002Fdemo1-voice-conversation.gif\" alt=\"Voice Conversation Demo\" width=\"100%\">\n\u003C\u002Fa>\n\u003Cbr>\u003Csub>Click for full video with audio\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003Ctd width=\"50%\" align=\"center\">\n\u003Cstrong>App Control\u003C\u002Fstrong>\u003Cbr>\n\u003Cem>Control Spotify, adjust volume — 38 macOS actions by voice.\u003C\u002Fem>\u003Cbr>\u003Cbr>\n\u003Ca href=\"https:\u002F\u002Fyoutu.be\u002FeTYwkgNoaKg\">\n\u003Cimg src=\"assets\u002Fdemos\u002Fdemo2-spotify-volume.gif\" alt=\"App Control Demo\" width=\"100%\">\n\u003C\u002Fa>\n\u003Cbr>\u003Csub>Click for full video with audio\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd width=\"50%\" align=\"center\">\n\u003Cstrong>Models\u003C\u002Fstrong>\u003Cbr>\n\u003Cem>Browse models, hot-swap LLMs — all from the TUI.\u003C\u002Fem>\u003Cbr>\u003Cbr>\n\u003Ca href=\"https:\u002F\u002Fyoutu.be\u002FHD1aS37zIGE\">\n\u003Cimg src=\"assets\u002Fdemos\u002Fdemo3-benchmarks.gif\" alt=\"Models & Benchmarks Demo\" width=\"100%\">\n\u003C\u002Fa>\n\u003Cbr>\u003Csub>Click for full video with audio\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003Ctd width=\"50%\" align=\"center\">\n\u003Cstrong>Document Intelligence (RAG)\u003C\u002Fstrong>\u003Cbr>\n\u003Cem>Ingest docs, ask questions by voice — ~4ms hybrid retrieval.\u003C\u002Fem>\u003Cbr>\u003Cbr>\n\u003Ca href=\"https:\u002F\u002Fyoutu.be\u002F8FEfbwS7cQ8\">\n\u003Cimg src=\"assets\u002Fdemos\u002Fdemo4-rag-documents.gif\" alt=\"RAG Demo\" width=\"100%\">\n\u003C\u002Fa>\n\u003Cbr>\u003Csub>Click for full video with audio\u003C\u002Fsub>\n\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n## Install\n\n> [IMPORTANT]\n> **Requires macOS 13+ on Apple Silicon. MetalRT engine requires M3 or later.** M1\u002FM2 Macs fall back to llama.cpp automatically.\n\n**One command:**\n\n```bash\ncurl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002FRunanywhereAI\u002FRCLI\u002Fmain\u002Finstall.sh | bash\n```\n\n**Or via Homebrew:**\n\n```bash\nbrew tap RunanywhereAI\u002Frcli https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI.git\nbrew install rcli\nrcli setup          # required — downloads AI models (~1GB, one-time)\n```\n\n**Upgrade to latest:**\n\n```bash\nbrew update\nbrew upgrade rcli\n```\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Troubleshooting: SHA256 mismatch or stale version\u003C\u002Fstrong>\u003C\u002Fsummary>\n\nIf `brew install` or `brew upgrade` fails with a checksum error:\n\n```bash\n# Force-refresh the tap to pick up the latest formula\ncd $(brew --repo RunanywhereAI\u002Frcli) && git fetch origin && git reset --hard origin\u002Fmain\nbrew reinstall rcli\n```\n\nIf that doesn't work, clean re-tap and clear the download cache:\n\n```bash\nbrew untap RunanywhereAI\u002Frcli\nrm -rf \"$(brew --cache)\u002Fdownloads\u002F\"*rcli*\nbrew tap RunanywhereAI\u002Frcli https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI.git\nbrew install rcli\nrcli setup\n```\n\n\u003C\u002Fdetails>\n\n## Quick Start\n\n```bash\nrcli                             # interactive TUI (push-to-talk + text)\nrcli listen                      # continuous voice mode\nrcli ask \"open Safari\"           # one-shot command\nrcli ask \"play some jazz on Spotify\"\nrcli vlm photo.jpg \"what's in this image?\"  # vision analysis\nrcli camera                      # live camera VLM\nrcli screen                      # screen capture VLM\nrcli metalrt                     # MetalRT GPU engine management\nrcli llamacpp                    # llama.cpp engine management\n```\n\n\n## Benchmarks\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fdecode-vs-llamacpp.webp\" alt=\"MetalRT vs llama.cpp decode speed\" width=\"700\" \u002F>\n  \u003Cbr>\n  \u003Cem>MetalRT decode throughput vs llama.cpp and Apple MLX on Apple M3 Max\u003C\u002Fem>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Frtf_comparison.webp\" alt=\"STT and TTS real-time factor comparison\" width=\"700\" \u002F>\n  \u003Cbr>\n  \u003Cem>STT and TTS real-time factor — lower is better. MetalRT STT is 714x faster than real-time.\u003C\u002Fem>\n\u003C\u002Fp>\n\nFor More info : \n- https:\u002F\u002Fwww.runanywhere.ai\u002Fblog\u002Fmetalrt-fastest-llm-decode-engine-apple-silicon\n- https:\u002F\u002Fwww.runanywhere.ai\u002Fblog\u002Fmetalrt-speech-fastest-stt-tts-apple-silicon\n- https:\u002F\u002Fwww.runanywhere.ai\u002Fblog\u002Ffastvoice-on-device-voice-ai-pipeline-apple-silicon\n\n## Features\n\n### Voice Pipeline\n\nA full STT + LLM + TTS pipeline running on Metal GPU with three concurrent threads:\n\n- **VAD** — Silero voice activity detection\n- **STT** — Zipformer streaming + Whisper \u002F Parakeet offline\n- **LLM** — Qwen3 \u002F LFM2 \u002F Qwen3.5 with KV cache continuation and Flash Attention\n- **TTS** — Double-buffered sentence-level synthesis (next sentence renders while current plays)\n- **Tool Calling** — LLM-native tool call formats (Qwen3, LFM2, etc.)\n- **Multi-turn Memory** — Sliding window conversation history with token-budget trimming\n\n### Vision (VLM)\n\nAnalyze images, camera captures, and screen regions using on-device vision-language models. VLM runs on the llama.cpp engine via Metal GPU — no cloud.\n\n- **Image Analysis** — `rcli vlm photo.jpg \"describe this\"` for single-image queries\n- **Camera** — Press **V** in the TUI or run `rcli camera` for live camera analysis\n- **Screen Capture** — Press **S** in the TUI or run `rcli screen` to analyze screen regions\n- **Models** — Qwen3 VL 2B, Liquid LFM2 VL 1.6B, SmolVLM 500M — download on demand via `rcli models vlm`\n\n> **Note:** VLM is currently available on the llama.cpp engine. MetalRT VLM support is coming soon.\n\n### 40 macOS Actions\n\nControl your Mac by voice or text. The LLM routes intent to actions executed locally via AppleScript and shell commands.\n\n| Category | Examples |\n|----------|---------|\n| **Productivity** | `create_note`, `create_reminder`, `run_shortcut` |\n| **Communication** | `send_message`, `facetime_call` |\n| **Media** | `play_on_spotify`, `play_apple_music`, `play_pause`, `next_track`, `set_music_volume` |\n| **System** | `open_app`, `quit_app`, `set_volume`, `toggle_dark_mode`, `screenshot`, `lock_screen` |\n| **Web** | `search_web`, `search_youtube`, `open_url`, `open_maps` |\n\nRun `rcli actions` to see all 40, or toggle them on\u002Foff in the TUI Actions panel.\n\n> **Tip:** If tool calling feels unreliable, press **X** in the TUI to clear the conversation and reset context. With small LLMs, accumulated context can degrade tool-calling accuracy — a fresh context often fixes it.\n\n### RAG (Local Document Q&A)\n\nIndex local documents, query them by voice. Hybrid vector + BM25 retrieval with ~4ms latency over 5K+ chunks. Supports PDF, DOCX, and plain text.\n\n```bash\nrcli rag ingest ~\u002FDocuments\u002Fnotes\nrcli ask --rag ~\u002FLibrary\u002FRCLI\u002Findex \"summarize the project plan\"\n```\n\n### Interactive TUI\n\nA terminal dashboard with push-to-talk, live hardware monitoring, model management, and an actions browser.\n\n| Key | Action |\n|-----|--------|\n| **SPACE** | Push-to-talk |\n| **V** | Camera — capture and analyze with VLM |\n| **S** | Screen — capture and analyze a screen region with VLM |\n| **M** | Models — browse, download, hot-swap LLM\u002FSTT\u002FTTS\u002FVLM |\n| **A** | Actions — browse, enable\u002Fdisable macOS actions |\n| **R** | RAG — ingest documents |\n| **X** | Clear conversation and reset context |\n| **T** | Toggle tool call trace |\n| **ESC** | Stop \u002F close \u002F quit |\n\n## MetalRT GPU Engine\n\nMetalRT is a high-performance GPU inference engine built by [RunAnywhere, Inc.](https:\u002F\u002Frunanywhere.ai) specifically for Apple Silicon. It delivers the fastest on-device inference for LLM, STT, and TTS — up to **550 tok\u002Fs** LLM throughput and sub-200ms end-to-end voice latency.\n\n> **Apple M3 or later required.** MetalRT uses Metal 3.1 GPU features available on M3, M3 Pro, M3 Max, M4, and later chips. M1\u002FM2 support is coming soon. On M1\u002FM2, RCLI automatically falls back to the open-source llama.cpp engine.\n\nMetalRT is automatically installed during `rcli setup` (choose \"MetalRT\" or \"Both\"). Or install separately:\n\n```bash\nrcli metalrt install\nrcli metalrt status\n```\n\n**Supported models:** Qwen3 0.6B, Qwen3 4B, Llama 3.2 3B, LFM2.5 1.2B (LLM) · Whisper Tiny\u002FSmall\u002FMedium (STT) · Kokoro 82M with 28 voices (TTS)\n\nMetalRT is distributed under a [proprietary license](https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002Fmetalrt-binaries\u002Fblob\u002Fmain\u002FLICENSE). For licensing inquiries: founder@runanywhere.ai\n\n## Supported Models\n\nRCLI supports 20+ models across LLM, STT, TTS, VLM, VAD, and embeddings. All run locally on Apple Silicon. Use `rcli models` to browse, download, or switch.\n\n**LLM:** LFM2 1.2B (default), LFM2 350M, LFM2.5 1.2B, LFM2 2.6B, Qwen3 0.6B, Qwen3.5 0.8B\u002F2B\u002F4B, Qwen3 4B\n\n**STT:** Zipformer (streaming), Whisper base.en (offline, default), Parakeet TDT 0.6B (~1.9% WER)\n\n**TTS:** Piper Lessac\u002FAmy, KittenTTS Nano, Matcha LJSpeech, Kokoro English\u002FMulti-lang\n\n**VLM:** Qwen3 VL 2B, Liquid LFM2 VL 1.6B, SmolVLM 500M — on-demand download via `rcli models vlm` (llama.cpp engine only)\n\n**Default install** (`rcli setup`): ~1GB — LFM2 1.2B + Whisper + Piper + Silero VAD + Snowflake embeddings. VLM models are downloaded on demand.\n\n```bash\nrcli models                  # interactive model management\nrcli models vlm              # download\u002Fmanage VLM models\nrcli upgrade-llm             # guided LLM upgrade\nrcli voices                  # browse and switch TTS voices\nrcli cleanup                 # remove unused models\n```\n\n## Build from Source\n\nCPU-only build using llama.cpp + sherpa-onnx (no MetalRT):\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002FRCLI.git && cd RCLI\nbash scripts\u002Fsetup.sh\nbash scripts\u002Fdownload_models.sh\nmkdir -p build && cd build\ncmake .. -DCMAKE_BUILD_TYPE=Release\ncmake --build . -j$(sysctl -n hw.ncpu)\n.\u002Frcli\n```\n\nAll dependencies are vendored or CMake-fetched. Requires CMake 3.15+ and Apple Clang (C++17).\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>CLI Reference\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```\nrcli                          Interactive TUI (push-to-talk + text + trace)\nrcli listen                   Continuous voice mode\nrcli ask \u003Ctext>               One-shot text command\nrcli vlm \u003Cimage> [prompt]     Analyze an image with VLM\nrcli camera [prompt]          Live camera capture + VLM analysis\nrcli screen [prompt]          Screen capture + VLM analysis\nrcli actions [name]           List actions or show detail\nrcli rag ingest \u003Cdir>         Index documents for RAG\nrcli rag query \u003Ctext>         Query indexed documents\nrcli models [llm|stt|tts|vlm] Manage AI models\nrcli voices                   Manage TTS voices\nrcli metalrt                  MetalRT GPU engine management\nrcli llamacpp                 llama.cpp engine management\nrcli setup                    Download default models\nrcli info                     Show engine and model info\n\nOptions:\n  --models \u003Cdir>      Models directory (default: ~\u002FLibrary\u002FRCLI\u002Fmodels)\n  --rag \u003Cindex>       Load RAG index for document-grounded answers\n  --gpu-layers \u003Cn>    GPU layers for LLM (default: 99 = all)\n  --ctx-size \u003Cn>      LLM context size (default: 4096)\n  --no-speak          Text output only (no TTS)\n  --verbose, -v       Debug logs\n```\n\n\u003C\u002Fdetails>\n\n## Contributing\n\nContributions welcome. See [CONTRIBUTING.md](CONTRIBUTING.md) for build instructions and how to add new actions, models, or voices.\n\n## License\n\nRCLI is open source under the [MIT License](LICENSE).\n\nMetalRT is proprietary software by [RunAnywhere, Inc.](https:\u002F\u002Frunanywhere.ai), distributed under a separate [license](https:\u002F\u002Fgithub.com\u002FRunanywhereAI\u002Fmetalrt-binaries\u002Fblob\u002Fmain\u002FLICENSE).\n\n\u003Cp align=\"center\">\n  Built by \u003Ca href=\"https:\u002F\u002Fwww.runanywhere.ai\">RunAnywhere, Inc.\u003C\u002Fa>\n\u003C\u002Fp>\n","RCLI 是一个针对 macOS 的本地语音AI助手，无需云端支持即可实现与Mac对话和文档查询。该项目集成了从语音识别到文本生成的完整流程，包括STT（语音转文字）、LLM（大语言模型）、TTS（文字转语音）以及VLM（视觉语言模型），能够执行40种macOS操作，并通过本地RAG技术对个人文档进行快速检索，响应时间低于200毫秒。其核心技术基于专为Apple Silicon芯片设计的MetalRT GPU推理引擎。RCLI适用于需要在本地环境中利用AI增强工作效率和个人体验的场景，尤其是对于重视数据隐私且偏好离线工作的用户来说非常理想。","2026-06-11 03:52:30","high_star"]