🟢🟢🟢🟢🟢 5. techjarves/Portable-Local-Studio
解压就能用的本地 AI 工作室。
一个界面里画图、跟本地大模型聊天、语音转文字、文字转语音，全程离线，不要 API Key，自动适配 N 卡/A 卡/Mac。
🔗 直达：techjarves/Portable-Local-Studio
以前四个工具四份订阅，现在一个文件夹全搞定，73 星一天！
10 views
offline ai from pendrive
techjarves/Portable-AI-USB: A 100% offline, fully portable, zero-trace AI (Ollama + Llama 3 + AnythingLLM) that runs natively from a USB drive on Windows and Mac.
techjarves/Portable-AI-USB
42 views
DroidDesk: Run a full Linux desktop on any Android phone (3.9k stars)
What it does
All text below is quoted directly from the project README.
>Run a full Linux desktop on any Android phone. Not a terminal. Not an emulator. A complete desktop environment with direct kernel access -- VS Code, Blender, Metasploit, local AI, all of it.
>Connect your phone to a monitor and it becomes a Linux PC. Unplug it and your entire setup comes with you.
>DroidDesk is an independent GPL-3.0 open-source project that incorporates modified Termux:X11 components. It is not affiliated with or endorsed by Termux, Termux:X11, TUR, Canonical, or Ubuntu.
What This Actually Runs:
>Everything below has been tested and confirmed working:
- LibreOffice -- Word processing, spreadsheets, presentations. Fully functional.
- VS Code -- Full version. Python, PIP, extensions, everything.
- Claude Code -- AI coding agent running directly in terminal.
- Blender -- Installs and opens. Laggy on mobile hardware, but it runs.
- Wireshark -- Full network analysis, every packet and protocol.
- Metasploit -- Pentesting framework, runs fine.
- Local AI -- Offline LLM inference, 5+ tokens/second, no API needed.
>If it runs on Ubuntu, it runs here.
How It Works:
>The Linux environment runs through Termux with direct access to the phone's kernel. No emulation, no translation -- native performance.
>The setup script installs a full desktop (XFCE4/LXQt/MATE/KDE) inside Termux using the Termux User Repository (TUR) for GUI apps. For tools not available in TUR (Wireshark, Metasploit, etc.), a Proot container provides a standard Ubuntu/Debian/Kali environment where you install anything with apt.
>The automatic menu sync scans what you install inside Proot and adds it directly to your desktop app menu. No need to enter the container every time.
DroidDesk App (Standalone):
>DroidDesk is also available as a standalone Android application that completely automates this process without requiring a separate Termux installation. It renders through an embedded Termux:X11 server running in its own Android process; the app does not use VNC.
- Rooted phones: Run the Ubuntu filesystem through chroot.
- Non-rooted phones: Run an app-private native Termux userspace and install desktop packages from the X11 and TUR repositories. PRoot is not used.
- Rendering: Both modes connect directly to the embedded X11 server on DISPLAY=:0. Adreno devices use Turnip/Zink hardware acceleration when available; other GPUs fall back to Mesa software rendering.
- Automated setup: The app extracts the bundled ARM64 Termux bootstrap, configures its private package prefix, and installs the selected desktop automatically.
>Download the latest release APK from the Releases tab and sideload it to begin.
Requirements:
- Any Android phone (ARM64)
- Termux (install from F-Droid, not Play Store)
- Termux-X11 (for on-phone display)
Repository
orailnoor/DroidDesk
Repository stats
|Metric|Value|
|:-|:-|
|Stars|3,909|
|Forks|373|
|Last release|v1.3.1 (2026-10-14)|
|Languages|Kotlin 43.3%, Dart 30.8%, Java 17.7%|
|Contributors|2|
|AI contributor accounts (Claude / Codex / Cursor / Copilot)|None|
|AI co-authored commits|None|
Main contributor
|Field|Value|
|:-|:-|
|Name|Tech Jarves|
|GitHub|@techjarves|
|Commits to this repo|31 (top contributor)|
|Bio|I break, tweak, and rebuild Android & Linux systems.|
|X / Twitter|@jarves_usaram|
|Followers|1.6k|
312 points · 11 comments
Run Uncensored GGUF LLMs Offline on Android, iOS, and Desktop with Flutter
Run Uncensored GGUF LLMs Offline on Android, iOS, and Desktop with Flutter
Deploying on-device AI used to mean writing complex platform-specific bindings across C++, Swift, and Kotlin. techjarves/Uncensored-Local-AI-Multiplatform breaks that barrier by delivering a unified Flutter app that runs quantized GGUF models directly on edge devices—completely offline, unmetered, and uncensored.
---
TL;DR
This project eliminates cloud API fees and data-privacy risks by running GGUF-based open-source LLMs 100% on-device across Android, iOS, and PC. Built on Flutter and powered by on-device inference engines (llama.cpp bindings), it bypasses cloud dependencies and guardrails, giving developers a turnkey blueprint for cross-platform local AI.
---
Key Technical Highlights
- Zero Cloud Reliance: Inference runs entirely locally. Zero telemetry, zero external API latency, and complete privacy.
- Universal GGUF Compatibility: Load popular small-footprint quantized models (Llama 3 8B, Mistral 7B, Phi-3, Qwen 2.5) directly into system memory.
- Cross-Platform Single Codebase: Written in Dart with Flutter, abstracting platform-specific acceleration APIs across macOS, Windows, Linux, iOS, and Android.
- Native Hardware Acceleration: Hooks into device capabilities (Metal on iOS/macOS, Vulkan/OpenCL/NNAPI on Android/Linux) via underlying C/C++ bindings for optimal tokens-per-second (TPS).
- Uncensored by Default: Designed to run raw weights without proprietary intermediary moderation layers, ideal for red-teaming, uncensored creative tasks, and specialized domain prompts.
---
Quick Start
Make sure you have Flutter installed. Clone the repository and install the dependencies:
# 1. Clone the repository
git clone https://github.com/techjarves/Uncensored-Local-AI-Multiplatform.git
cd Uncensored-Local-AI-Multiplatform
# 2. Fetch Flutter packages
flutter pub get
# 3. Run on your target device (e.g., Desktop or connected mobile)
flutter run -d macos # or windows, linux, android, ios
Once launched, transfer your preferred quantized model (e.g., mistral-7b-instruct-v0.2.Q4_K_M.gguf) into the app's designated model storage directory or use the integrated file picker to begin instant local generation.
---
Why It Matters
- Edge AI Engineers: Provides a battle-tested reference architecture for embedding high-performance local inference in Flutter applications.
- Privacy-First Developers: Ideal for healthcare, legal, or sensitive enterprise applications where data must never leave the physical device boundary.
- Offline First Products: Essential for tools requiring zero-latency offline text generation in disconnected environments (fieldwork, aviation, remote infrastructure).
---
🛠️ Recommended AI Stack & Resources
- Cloud GPU Hosting: Need to fine-tune your GGUF models or test larger 70B parameter checkpoints before deploying to edge? Spin up low-cost on-demand GPUs with RunPod.
- AI Code Editor: Build Flutter apps and integrate C++ bindings 10x faster with Cursor.
- Production Database: Add vector search and long-term memory to your local agents using Supabase or Pinecone.
- Global Dev Network: Ensure low-latency, reliable model weight downloads from Hugging Face and fast cloud server synchronization with WD-Gold Network.
- Newsletter: Stay ahead of the curve in edge intelligence. Subscribe to Local AI Daily for curated breakdowns of on-device LLMs, inference runtimes, and open-source models.
- Sponsorship & Partnerships: Want to feature your developer tool or AI infrastructure project to our engineering audience? Contact: [fengdahui195@gmail.com](mailto:fengdahui195@gmail.com).
This guy went ahead and built what I was talking about: techjarves/Mobile-Harness
Mentions
🟢🟢🟢🟢🟢 5. techjarves/Portable-Local-Studio 解压就能用的本地 AI 工作室。 一个界面里画图、跟本地大模型聊天、语音转文字、文字转语音，全程离线，不要 API Key，自动适配 N 卡/A 卡/Mac。 🔗 直达：techjarves/Portable-Local-Studio 以前四个工具四份订阅，现在一个文件夹全搞定，73 星一天！
offline ai from pendrive techjarves/Portable-AI-USB: A 100% offline, fully portable, zero-trace AI (Ollama + Llama 3 + AnythingLLM) that runs natively from a USB drive on Windows and Mac. techjarves/Portable-AI-USB
DroidDesk: Run a full Linux desktop on any Android phone (3.9k stars) What it does All text below is quoted directly from the project README. >Run a full Linux desktop on any Android phone. Not a terminal. Not an emulator. A complete desktop environment with direct kernel access -- VS Code, Blender, Metasploit, local AI, all of it. >Connect your phone to a monitor and it becomes a Linux PC. Unplug it and your entire setup comes with you. >DroidDesk is an independent GPL-3.0 open-source project that incorporates modified Termux:X11 components. It is not affiliated with or endorsed by Termux, Termux:X11, TUR, Canonical, or Ubuntu. What This Actually Runs: >Everything below has been tested and confirmed working: - LibreOffice -- Word processing, spreadsheets, presentations. Fully functional. - VS Code -- Full version. Python, PIP, extensions, everything. - Claude Code -- AI coding agent running directly in terminal. - Blender -- Installs and opens. Laggy on mobile hardware, but it runs. - Wireshark -- Full network analysis, every packet and protocol. - Metasploit -- Pentesting framework, runs fine. - Local AI -- Offline LLM inference, 5+ tokens/second, no API needed. >If it runs on Ubuntu, it runs here. How It Works: >The Linux environment runs through Termux with direct access to the phone's kernel. No emulation, no translation -- native performance. >The setup script installs a full desktop (XFCE4/LXQt/MATE/KDE) inside Termux using the Termux User Repository (TUR) for GUI apps. For tools not available in TUR (Wireshark, Metasploit, etc.), a Proot container provides a standard Ubuntu/Debian/Kali environment where you install anything with apt. >The automatic menu sync scans what you install inside Proot and adds it directly to your desktop app menu. No need to enter the container every time. DroidDesk App (Standalone): >DroidDesk is also available as a standalone Android application that completely automates this process without requiring a separate Termux installation. It renders through an embedded Termux:X11 server running in its own Android process; the app does not use VNC. - Rooted phones: Run the Ubuntu filesystem through chroot. - Non-rooted phones: Run an app-private native Termux userspace and install desktop packages from the X11 and TUR repositories. PRoot is not used. - Rendering: Both modes connect directly to the embedded X11 server on DISPLAY=:0. Adreno devices use Turnip/Zink hardware acceleration when available; other GPUs fall back to Mesa software rendering. - Automated setup: The app extracts the bundled ARM64 Termux bootstrap, configures its private package prefix, and installs the selected desktop automatically. >Download the latest release APK from the Releases tab and sideload it to begin. Requirements: - Any Android phone (ARM64) - Termux (install from F-Droid, not Play Store) - Termux-X11 (for on-phone display) Repository orailnoor/DroidDesk Repository stats |Metric|Value| |:-|:-| |Stars|3,909| |Forks|373| |Last release|v1.3.1 (2026-10-14)| |Languages|Kotlin 43.3%, Dart 30.8%, Java 17.7%| |Contributors|2| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|None| |AI co-authored commits|None| Main contributor |Field|Value| |:-|:-| |Name|Tech Jarves| |GitHub|@techjarves| |Commits to this repo|31 (top contributor)| |Bio|I break, tweak, and rebuild Android & Linux systems.| |X / Twitter|@jarves_usaram| |Followers|1.6k|
Run Uncensored GGUF LLMs Offline on Android, iOS, and Desktop with Flutter Run Uncensored GGUF LLMs Offline on Android, iOS, and Desktop with Flutter Deploying on-device AI used to mean writing complex platform-specific bindings across C++, Swift, and Kotlin. techjarves/Uncensored-Local-AI-Multiplatform breaks that barrier by delivering a unified Flutter app that runs quantized GGUF models directly on edge devices—completely offline, unmetered, and uncensored. --- TL;DR This project eliminates cloud API fees and data-privacy risks by running GGUF-based open-source LLMs 100% on-device across Android, iOS, and PC. Built on Flutter and powered by on-device inference engines (llama.cpp bindings), it bypasses cloud dependencies and guardrails, giving developers a turnkey blueprint for cross-platform local AI. --- Key Technical Highlights - Zero Cloud Reliance: Inference runs entirely locally. Zero telemetry, zero external API latency, and complete privacy. - Universal GGUF Compatibility: Load popular small-footprint quantized models (Llama 3 8B, Mistral 7B, Phi-3, Qwen 2.5) directly into system memory. - Cross-Platform Single Codebase: Written in Dart with Flutter, abstracting platform-specific acceleration APIs across macOS, Windows, Linux, iOS, and Android. - Native Hardware Acceleration: Hooks into device capabilities (Metal on iOS/macOS, Vulkan/OpenCL/NNAPI on Android/Linux) via underlying C/C++ bindings for optimal tokens-per-second (TPS). - Uncensored by Default: Designed to run raw weights without proprietary intermediary moderation layers, ideal for red-teaming, uncensored creative tasks, and specialized domain prompts. --- Quick Start Make sure you have Flutter installed. Clone the repository and install the dependencies: # 1. Clone the repository git clone https://github.com/techjarves/Uncensored-Local-AI-Multiplatform.git cd Uncensored-Local-AI-Multiplatform # 2. Fetch Flutter packages flutter pub get # 3. Run on your target device (e.g., Desktop or connected mobile) flutter run -d macos # or windows, linux, android, ios Once launched, transfer your preferred quantized model (e.g., mistral-7b-instruct-v0.2.Q4_K_M.gguf) into the app's designated model storage directory or use the integrated file picker to begin instant local generation. --- Why It Matters - Edge AI Engineers: Provides a battle-tested reference architecture for embedding high-performance local inference in Flutter applications. - Privacy-First Developers: Ideal for healthcare, legal, or sensitive enterprise applications where data must never leave the physical device boundary. - Offline First Products: Essential for tools requiring zero-latency offline text generation in disconnected environments (fieldwork, aviation, remote infrastructure). --- 🛠️ Recommended AI Stack & Resources - Cloud GPU Hosting: Need to fine-tune your GGUF models or test larger 70B parameter checkpoints before deploying to edge? Spin up low-cost on-demand GPUs with RunPod. - AI Code Editor: Build Flutter apps and integrate C++ bindings 10x faster with Cursor. - Production Database: Add vector search and long-term memory to your local agents using Supabase or Pinecone. - Global Dev Network: Ensure low-latency, reliable model weight downloads from Hugging Face and fast cloud server synchronization with WD-Gold Network. - Newsletter: Stay ahead of the curve in edge intelligence. Subscribe to Local AI Daily for curated breakdowns of on-device LLMs, inference runtimes, and open-source models. - Sponsorship & Partnerships: Want to feature your developer tool or AI infrastructure project to our engineering audience? Contact: [fengdahui195@gmail.com](mailto:fengdahui195@gmail.com).
This guy went ahead and built what I was talking about: techjarves/Mobile-Harness