Run local LLMs, generate photorealistic Cinema SDXL art, analyze complex spreadsheets, build interactive courses, snip screens, and chat by voice β Local-First by default, Air-Gapped capable, with Zero silent network calls.
π Official Website β’ π¦ Download v1.1.0 β’ β¨ Feature Tour β’ π Benchmarks β’ π οΈ Developer Guide β’ π’ Enterprise PRO β’ π€ Contributing
Welcome to the official open-source repository of AiWidget!
This repository contains the complete client workspace:
- π» 100% Frontend Source Code (
src/): High-performance reactive TypeScript 5.4, Vite, Web Audio,$\KaTeX$ math engine, Mermaid.js diagramming, DOM sanitization, and floating widget shells. - π¦ 100% Embedded Native Rust Engine (
src-tauri/): High-speed C++sd.cppdiffusion wrapper, Whisper.cpp speech recognition, neural text-to-speech, in-memory streaming Excel/PDF RAG parsers, encrypted SQLite database, and mutual cryptographic transport.
Anyone in the worldwide developer community can clone this repository, run npm run tauri:dev, inspect the code, add new AI engines, translate into new languages, optimize performance, and compile standalone binaries.
flowchart LR
subgraph OpenSource ["π Public Repository (This Repo - GNU AGPLv3)"]
UI["TypeScript Frontend (src/)"]
RUST["Embedded Native Rust Engine (src-tauri/)"]
ENG["Local AI Engines (Ollama, SD.cpp, Whisper, SQLite)"]
UI <--> RUST <--> ENG
end
subgraph PrivateHub ["π’ Private Enterprise Infrastructure"]
SERVER["On-Premise Multi-User Server (Central GPU Pool)"]
LICENSE["HMAC License & Telemetry Console"]
end
RUST -.->|"Mutual Signed Cryptographic Protocol"| SERVER
Today, developers, professionals, and privacy-conscious users face two frustrating compromises:
- Cloud AI Privacy Risks (Microsoft Copilot, ChatGPT Desktop): Every prompt, proprietary source code snippet, financial spreadsheet (.xlsx), medical record, and screen snip is transmitted to third-party cloud servers. For regulated industries (lawyers, CPAs, doctors, defense, engineering), this violates GDPR, HIPAA, and legal client privilege.
- Bloated Electron Wrappers: Traditional open-source desktop clients consume 500 MB to 1 GB of RAM just to display a chat window, take up hundreds of megabytes on disk, and lack integrated image studios, document parsers, or course generators.
Engineered natively in Rust & Tauri, AiWidget delivers a complete multimodal AI workstation in a discreet floating desktop widget:
- πͺΆ Featherweight Footprint: 9.03 MB installer size, < 80 MB resident RAM footprint.
- π‘οΈ Zero-Knowledge Air-Gapped Privacy: All LLM inference, OCR parsing, and database queries run 100% on your machine.
- π§© True Multimodal Independence: Local LLMs, Cinema SDXL photorealistic rendering, spreadsheet financial extraction, course generation, screen snipping, and Whisper voice interaction β unified in one lightweight native app.
| Capability | AiWidget | LM Studio | Jan.ai | AnythingLLM | Microsoft Copilot |
|---|---|---|---|---|---|
| Form Factor | π’ Floating Widget / Bar / Popup | π΄ Full Window | π΄ Full Window | π‘ Partial Overlay | π’ Edge Sidebar |
| Installer Size | π’ 9.03 MB | π΄ ~350 MB | π΄ ~160 MB | π΄ ~380 MB | π΄ Heavy Webview |
| RAM Consumption | π’ < 80 MB | π‘ ~400 MB | π΄ ~450 MB | π΄ ~550 MB | π΄ Cloud / Webview |
| Data Sovereignty | π’ 100% Local (Air-Gapped) | π’ 100% Local | π’ 100% Local | π’ 100% Local | π΄ Cloud Tracking |
| Cinema SDXL Studio (6.6 GB) | π’ Built-in + CPU VAE Tiling | π΄ No | π΄ No | π΄ No | π‘ Cloud DALL-E (Paid) |
| Fast SD 1.5 Studio (1.5 GB) | π’ Built-in (15s renders) | π΄ No | π΄ No | π΄ No | π΄ No |
| Spreadsheet Intelligence (.xlsx) | π’ Native Extraction & Sums | π΄ No | π΄ No | π‘ Text only | π‘ Cloud limited |
| Course & Training Studio | π’ Curriculum, Quiz, Word (.docx) | π΄ No | π΄ No | π΄ No | π΄ No |
| 1-Click Screen Snipper | π’ Ctrl+Shift+S + OCR |
π΄ No | π΄ No | π΄ No | π’ Cloud Vision |
| Whisper Voice + Neural TTS | π’ Continuous Loop | π΄ No | π΄ No | π΄ No | π’ Cloud |
| Anti-Hallucination Web Router | π’ Auto-Trigger on Demand | π΄ No | π΄ No | π‘ Paid API Key | π’ Bing Cloud |
| Interactive Mermaid & LaTeX | π’ Zoom, Fullscreen & Export | π‘ Basic code | π΄ Raw text | π‘ Basic | π‘ Basic |
| Enterprise Server & Quotas | π’ On-Premise Ready | π΄ No | π΄ No | π‘ Basic | π΄ 30$/user/month |
AiWidget is fundamentally local-first, privacy-driven, and model-agnostic:
- π§© Universal Local Model Support: It natively connects to any local LLM runtime via Ollama, Llama.cpp, and local GGUF models (including Qwen 2.5, Llama 3.3, Mistral, DeepSeek-R1, Gemma, and Phi-3).
- β‘ Hardware-Proportional Intelligence: The analytical depth, reasoning power, and token speed scale directly with your chosen local model and hardware (from lightweight 3B/7B models on standard laptops to 14B/32B/70B models on high-performance GPU workstations).
- π 100% Open-Source & Auditable Integrity: No mandatory cloud subscriptions, no black-box API dependencies, and no hidden tracking. 100% of the desktop client source code (TypeScript + Rust) is available in this repository under the GNU AGPLv3 for full community audit and verification.
flowchart TD
subgraph Core ["AiWidget All-In-One Local Hub"]
LLM["π§ Local LLMs (Ollama / Llama.cpp)"]
SD["π¨ Cinema SDXL + SD 1.5 Image Studio"]
RAG["π Excel & PDF Financial Analyzer"]
CRS["π Interactive Course & Quiz Studio"]
SNP["πΈ 1-Click Screen Snipper (Ctrl+Shift+S)"]
VCE["ποΈ Whisper Voice Chat & Neural TTS"]
WEB["π On-Demand Anti-Hallucination Web Router"]
SEC["π Multi-Profile Encrypted SQLite"]
end
Forget setting up complex 30 GB Python environments or paying monthly cloud subscriptions. AiWidget embeds a high-performance C++ (sd.cpp) diffusion pipeline directly in Rust:
- π¬ Cinema SDXL β Juggernaut XL v8 (6.6 GB):
- Generates hyper-photorealistic 1024x1024 cinema-quality artwork, character portraits, product concepts, and architectural visualizations.
- Automated Prompt Translation & Expansion: Type prompts in French or Arabic; AiWidget automatically translates and optimizes them into specialized Stable Diffusion positive and negative conditioning tokens (inspired by Fooocus-style cinematic presets).
- Built-in Cinematic Styles: Photorealistic 8K, Masterpiece, Cinematic Lighting, Hyper-Detailed Skin Texture, and Anime/Digital Art presets.
- β‘ Rapid SD 1.5 (1.5 GB):
- Ultra-fast generation (15β20 seconds) engineered for standard consumer laptops, older GPUs, and quick ideation.
- π‘οΈ Smart CPU VAE Tiling Fallback (Zero OOM Crashes):
- Traditional SDXL software crashes with
CUDA Out-Of-Memoryon 4 GB or 6 GB VRAM GPUs during the heavy VAE decode step. - AiWidget automatically detects low VRAM thresholds and offloads VAE tiling computation to system RAM, guaranteeing successful renders without crashes.
- Traditional SDXL software crashes with
- πΌοΈ Interactive Image Gallery: Fullscreen modal view, instant PNG save, copy to clipboard, and generation metadata inspection (seed, steps, CFG, sampler).
A dedicated pedagogical engine designed for teachers, corporate instructors, university students, and self-learners:
- π Complete Curriculum Structuring: Generates structured courses tailored to any difficulty level (Beginner, Intermediate, Advanced), including learning objectives, prerequisite summaries, and module breakdowns.
- π‘ Step-by-Step Lessons with Real-World Analogies: Breaks down abstract technical, legal, scientific, or business topics into digestible lessons illustrated with practical case studies.
- π Interactive Quizzes & MCQs: Automatically embeds multiple-choice questions with detailed answer keys, feedback explanations, and scoring mechanisms.
- π Universal 1-Click Export:
- Export directly to formatted Microsoft Word (.docx) with custom heading hierarchies, callout boxes, and styled tables.
- Export to clean Markdown (.md) for seamless integration into Obsidian, Notion, or GitHub.
Reduce hallucinations when fresh factual knowledge is required, while strictly preserving your privacy:
- π Local-First by Default β Zero Silent Network Requests:
- All core intelligence, LLM inference, image generation, voice chat, and document RAG run 100% locally on your device.
- Internet access is never activated silently.
- π€ Smart Intent Detection & Explicit Permission Gate:
- When you ask a question requiring live or time-sensitive data (e.g., "What will the weather be tomorrow in New York?", "Latest stock market indices", or "Recent software releases"), AiWidget automatically detects the intent.
- Instead of quietly calling the network, AiWidget displays an explicit confirmation gate:
π This question requires live web information. Allow AiWidget to search the web? [Allow] [Deny] - You can also explicitly toggle the
π Webmode button in the chat input bar whenever you want web grounding enabled.
- π Real-Time Web Scraping & Fact Extraction:
- When authorized, AiWidget queries public search engines in the background without transmitting user identities.
- It strips HTML bloat, extracts relevant factual content, and injects retrieved web context into the LLM prompt.
- π Clickable Source Citations:
- Web-grounded responses include direct, clickable source URLs so you can verify the information directly from the cited sources.
- π‘οΈ Air-Gapped Capable:
- In strictly air-gapped or high-security corporate environments, Web Search can remain permanently disabled with zero impact on local LLM, SDXL, or document analysis features.
- π In-Memory Streaming XLSX & CSV Parser:
- Drag-and-drop spreadsheets, budget exports, or client estimates.
- Automatically reconstructs tabular columns, identifies currency rows, calculates sub-totals, tax amounts, and flags discrepancies.
- π Universal PDF & DOCX Extraction:
- Summarize multi-page legal contracts, research whitepapers, and invoices with zero cloud exposure.
- π OCR Image Extraction:
- Scanned invoices, receipts, and photos of paper documents are processed through our local OCR engine to extract editable text and numbers.
- β‘ Global Shortcut (
Ctrl + Shift + S) & Toolbar Camera Button[πΈ]:- Instantly snip any open window, software error code, compiler log, or web dashboard.
- π Native Clipboard Integration:
- Use standard Windows
Win + Shift + Sto grab an area, then pressCtrl + Vdirectly in AiWidget: the image thumbnail attaches automatically.
- Use standard Windows
- ποΈ Visual OCR & Multimodal Analysis:
- Prompts like "Explain this stack trace", "Translate this screenshot", or "Extract this table into Markdown" work immediately with local models.
- π€ Whisper.cpp Integration: High-accuracy local speech-to-text supporting multiple accents and languages.
- π£οΈ Neural Text-to-Speech: Crystal-clear male and female natural voices for English, French, and Arabic.
- π Continuous Hands-Free Conversation Mode: Speak naturally to your assistant, listen to its vocalized answer, and have it automatically resume listening for your follow-up questions.
- π
$\KaTeX$ Mathematical Engine: Render complex mathematical formulas, physics equations, matrices, and statistical distributions with crisp vector rendering. - π Mermaid.js Flowchart Engine: Visualizes flowcharts, architecture diagrams, sequence maps, class structures, and Gantt timelines with one-click full-screen expansion and PNG export.
- π₯ Isolated Workspaces: Create independent user profiles (e.g., Personal, Work, Client X) with dedicated conversation histories, custom system prompts, and individual master passwords.
- π Encrypted Local Storage: SQLite databases are stored under
%LOCALAPPDATA%\aiwidgetprotected with local cryptographic encryption.
Download the official signed releases from our Releases Page:
| Package | Format | Size | Target Environment |
|---|---|---|---|
| AI-Widget-Setup.exe | Installer | 9.03 MB | Standard Windows NSIS installer with Start Menu shortcuts & auto-update daemon. |
| AI-Widget-Setup.msi | MSI | 11.01 MB | Enterprise Windows Installer for automated Active Directory / GPO fleet deployment. |
| AI-Widget-Portable.exe | Portable | 21.55 MB | Zero-installation standalone executable. Ideal for USB drives and air-gapped computers. |
To build and develop AiWidget locally, ensure you have:
- Node.js (v18.0 or later) &
npm - Rust & Cargo (1.70 or later)
- C++ Build Tools for Visual Studio (MSVC x64)
# 1. Clone the repository
git clone https://github.com/ShaDevPro/AiWidget.git
cd AiWidget
# 2. Install frontend dependencies
npm install
# 3. Launch live hot-reload development mode
npm run tauri:dev# Compile frontend + Tauri binaries (Setup.exe, Setup.msi, Portable.exe)
npm run release:buildThe resulting optimized binaries will be placed in the release/ directory.
AiWidget/
βββ src/ # π» Full Frontend (TypeScript + Vite)
β βββ modules/
β β βββ chat/ # Streaming LLM chat controller & message renderer
β β βββ image/ # Cinema SDXL & Fast SD 1.5 Image Studio
β β βββ course/ # Interactive Course & Quiz Studio Engine
β β βββ snipper/ # ScreenSnipper (Ctrl+Shift+S screenshot tool)
β β βββ document/ # In-Memory Streaming Excel (.xlsx) & PDF parser
β β βββ voice/ # Whisper & Neural TTS Controller
β β βββ markdown/ # KaTeX math & Mermaid diagram renderers
β β βββ license/ # Enterprise server connection adapter
β β βββ telemetry/ # Privacy-compliant anonymous metrics pinger
β β βββ shell/ # Widget UI Shell (Bubble, Compact, Expanded)
β βββ styles/ # Modular Fluent Design CSS sheets
β βββ i18n/ # Multilingual dictionaries (FR, EN, AR RTL)
β
βββ src-tauri/ # π¦ Embedded Native Rust Backend Engine
β βββ src/
β β βββ sd_engine.rs # SD.cpp C++ diffusion wrapper (SDXL + CPU VAE)
β β βββ tts_engine.rs # Neural TTS speech synthesizer
β β βββ whisper_engine.rs # Whisper.cpp speech-to-text pipeline
β β βββ llama_engine.rs # Embedded Llama.cpp & Ollama IPC bridge
β β βββ web_search.rs # Anti-hallucination web query router
β β βββ db.rs # Local encrypted SQLite database engine
β β βββ security_engine.rs # Local hardware attestation & anti-tamper security engine
β β βββ commands/ # Tauri IPC Command Bridge (Async Handlers)
β βββ Cargo.toml # Rust dependencies & metadata
β βββ tauri.conf.json # Tauri window & bundle configuration
β
βββ website/ # π Public Showcase Website (GitHub Pages)
| Command | Description |
|---|---|
npm run tauri:dev |
Starts Vite dev server and launches the native Tauri desktop window with live hot-reload. |
npm run build |
Compiles and type-checks the TypeScript frontend (tsc && vite build). |
npm run tauri:build |
Compiles the optimized native Windows Rust desktop executable. |
npm run release:build |
Complete build: bundles frontend + builds Windows NSIS Installer, Enterprise MSI & Portable packages. |
AiWidget natively supports comprehensive bidirectional internationalization:
- π¬π§ English: Complete UI, prompt templates, and documentation.
- π«π· FranΓ§ais: Interface complΓ¨te, prompts pΓ©dagogiques et analyse de documents adaptΓ©e.
- πΈπ¦ Ψ§ΩΨΉΨ±Ψ¨ΩΨ©: Full Right-To-Left (RTL) typography, Arabic fonts, and custom Arabic LLM system directives.
Want to add a new language? Simply add a new locale file in src/i18n/locales/ and submit a Pull Request!
While the AiWidget Desktop Client is 100% free and open-source under the GNU AGPLv3, we offer an on-premise infrastructure for enterprises requiring centralized data sovereignty:
- π’ Centralized GPU Pooling: Connect 50 to 500 employee desktops to a central On-Premise GPU server without requiring discrete GPUs on client PCs.
- π₯ Department Quota Governance: Set fine-grained daily token limits for Engineering, Legal, and HR departments.
- π Active Directory & GPO Policy Lock: Enforce mandatory server endpoints and compliance rules across enterprise fleets.
- π‘οΈ Hardware-Locked HMAC Signature Licensing: Completely air-gapped license management with zero external cloud dependencies.
π§ Commercial & Enterprise Inquiries: s.h.a.dev.pro@gmail.com
π Website & Portal: https://shadevpro.github.io/AiWidget-Site/
We warmly welcome contributions from developers worldwide! Whether you want to optimize performance, add Linux/macOS support, translate the app into new languages, or introduce new document parsers:
- Fork the repository: https://github.com/ShaDevPro/AiWidget
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -m "feat: Add AmazingFeature") - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
This project is licensed under the GNU Affero General Public License v3.0 (AGPLv3).
See the LICENSE file for details.
Built with β€οΈ for privacy, performance, and the global developer community by ShaDevPro.
β If you find AiWidget valuable, please consider starring this repository on GitHub! β