RAGdoll

Local-first AI chat with retrieval-augmented generation.
Your conversations, your documents, your models — all on your machine.

---

What is RAGdoll?

RAGdoll is a desktop chat application that connects to the AI providers you already use while keeping every conversation and document stored locally. It automatically embeds your chat history and uploaded files into a private vector database, then retrieves the most relevant context for every message — no cloud, no telemetry, no subscriptions beyond your own API keys.

---

🆕 What's New — v0.2.0

📄 Document Creation (File R/W skill)

• WeasyPrint PDF engine — replaces reportlab as the primary PDF renderer; full CSS layout, custom fonts, page numbers, and header/footer via @page rules. reportlab remains as a fallback.
• Image embedding — the agent can now fetch images from the web and embed them directly into DOCX, PDF, and HTML documents.
• Smart text-wrap — images float left or right with text wrapping around them. Alignment is auto-detected from the image's aspect ratio (landscape → centre/full, portrait → float right) or controlled explicitly with ![caption|left](url), ![caption|right](url), ![caption|center](url), ![caption|full](url).
• DOCX floating images — uses native OOXML wp:anchor elements so Word honours the wrap style correctly.
• PPTX & XLSX support — python-pptx and openpyxl added for reading and writing PowerPoint and Excel files.

🖼️ Image Search plugin (new)

• Dedicated image-search skill — triggered only when the user explicitly asks for images or when the File R/W skill needs them for a document.
• Four providers: DuckDuckGo (default, no key), Unsplash, Pexels, Pixabay — switchable in plugin settings.
• Dynamic settings fields — API key inputs appear only for the selected provider (show_if conditional fields now supported across all plugin settings drawers).
• Pre-installed and auto-wired with the File R/W skill.

💬 Inline image display in chat

• Images returned by the agent are rendered as Discord-style thumbnails directly inside the chat bubble — loading skeleton, broken-URL fallback, and cursor-zoom-in hint.
• Click any image to open a full-screen lightbox with backdrop blur, smooth scale animation, and ESC / click-outside to dismiss.

🔄 Smarter update flow

• Two-phase download/install — clicking "Check for updates" no longer downloads immediately. A Download button appears first; progress, download speed (MB/s), bytes transferred, and ETA are shown inline while downloading. After download completes an Install & restart button appears.
• Same rich progress UI in both the top banner and the About settings tab.

🛠️ Other improvements

• /compact slash command — summarises and compresses long conversations to reclaim context window space.
• Context-overflow protection — the sidecar now detects token budget exhaustion and auto-compacts before hitting provider limits.
• Universal File Access skill — lets the agent read, search, and index files anywhere on disk.
• Context-menu permission dialogs for File R/W write/delete operations.
• Onboarding polish — correct repo links, white logo on the welcome screen, and a hidden Ctrl+Shift+F1 shortcut to replay onboarding.
• 13 RAG performance optimisations — ONNX-accelerated embedder, IVF-PQ index, prompt caching, LRU retrieval cache, semantic dedup, async HTTP pool, SSE compression, staged warm-up, and more.

---

✨ Features

🤖 Multi-provider	OpenAI, Anthropic, Groq, Google Gemini, Ollama, HuggingFace, OpenRouter
🧠 Persistent memory	Every message and document is embedded locally with ONNX-accelerated all-MiniLM-L6-v2
📄 Document ingestion	PDF, DOCX, TXT, CSV, JSON, PPTX, XLSX — drag-drop or attach inline
📝 Document creation	Generate DOCX, PDF, HTML from chat — with images, text-wrap, and styled layouts
🖼️ Image search	Find freely-licensed images (DDG / Unsplash / Pexels / Pixabay) and embed them in documents
🔍 Semantic search	Full memory browser with search, filtering, and per-chunk management
🧩 Plugins	Install community skills from the RAGdoll Plugins repo
👤 Model profiles	Named profiles per provider/model with separate API keys
🖥️ System tray	Hide to tray, launch on startup, new chat from tray menu
🔄 Smart updates	Background check; two-phase download/install with speed & ETA
🎨 Theming	Dark / light / system, custom accent colour, font size & density
⌨️ Keyboard-first	Ctrl+N new chat · Ctrl+P switch profile · Ctrl+, settings · and more

---

🧩 Plugins

RAGdoll's skill system is fully plugin-based. All community plugins live in the dedicated plugins repository:

github.com/itz-mune/RAGdoll-plugins

Pre-installed skills

Skill	Description
File R/W	Read, write, diff, and create rich documents (DOCX / PDF / HTML) with image embedding
Universal File Access	Index and search files anywhere on your filesystem
Image Search	Fetch freely-licensed images from the web; auto-wired with File R/W
Web Search	Live web search via DuckDuckGo / SearXNG
YouTube Transcripts	Pull transcripts from any YouTube video

Installing plugins

1. Open Settings → Plugins.
2. Browse the marketplace or paste a plugin URL.
3. Click Install — the skill is available immediately in the next chat.

Building your own plugin

Each plugin is a folder with a manifest.json (name, description, config fields) and a skill.py (a LangChain tool class). See the plugin authoring guide in the plugins repo for a full walkthrough.

---

📦 Installation

Download the latest installer for your platform from the Releases page:

Platform	File
Windows	RAGdoll_0.2.0_x64-setup.exe or _x64_en-US.msi
macOS (Apple Silicon)	RAGdoll_0.2.0_aarch64.dmg
macOS (Intel)	RAGdoll_0.2.0_x86_64.dmg
Linux	RAGdoll_0.2.0_amd64.AppImage or _amd64.deb

Latest release: v0.2.0

macOS: If Gatekeeper blocks the app, right-click → Open on first launch.

Windows: RAGdoll requires WebView2. Windows 11 and Windows 10 (post-2021) already include it — the installer will download it automatically on older systems.

Linux: RAGdoll requires libwebkit2gtk-4.1. Install it with:

sudo apt install libwebkit2gtk-4.1-0   # Ubuntu / Debian / Mint
sudo dnf install webkit2gtk4.1          # Fedora / RHEL

Alternatively, use the .AppImage which bundles most dependencies.

---

🚀 Build from source

Prerequisites

Tool	Version
Rust	stable
Node.js	20+
pnpm	9+
uv	latest

Windows (PDF support): WeasyPrint requires GTK3 runtime libraries. Install via the GTK3 for Windows runtime installer or via MSYS2:

pacman -S mingw-w64-x86_64-gtk3 mingw-w64-x86_64-pango mingw-w64-x86_64-cairo

Steps

# 1. Clone
git clone https://github.com/itz-mune/RAGdoll.git
cd RAGdoll

# 2. Install frontend dependencies
pnpm install

# 3. Install Python sidecar dependencies
cd sidecar && uv sync && cd ..

# 4. Run in development mode
pnpm tauri dev

The first run downloads and converts the embedding model to ONNX (~1 min, one-time).

Production build

pnpm tauri build

---

🏗️ Tech stack

Layer	Technology
Shell	Tauri v2 (Rust)
Frontend	React 19 · TypeScript · Vite · Tailwind CSS v4
Sidecar	Python · FastAPI · uvicorn
AI	LangChain · LangGraph
Memory	LanceDB · ONNX Runtime · all-MiniLM-L6-v2
State	Zustand · Tauri plugin-store
PDF	WeasyPrint (primary) · reportlab (fallback)
Documents	python-docx · python-pptx · openpyxl · Pillow
Packaging	PyInstaller · GitHub Actions (Win / macOS / Linux)

---

🤝 Contributing

Pull requests are welcome. For major changes please open an issue first.

# Type-check frontend
pnpm tsc --noEmit

# Lint Rust
cd src-tauri && cargo clippy

See RELEASING.md for the release workflow.

---

📄 License

MIT