Dazzi

Ask your files.

A Finder for your Mac that understands what's inside your documents, chats and mail. Dazzi powered by AutoRAG.

Website · Features · Develop · Architecture · Design · Releases

Features

🔎 Just ask

Ask the way you'd ask a teammate. Dazzi reads the files and writes back, with every answer linked to the passages behind it. Drag files, folders or links into a question to ask about just those.

🌻 Spot the twins

proposal-final, proposal-v3, proposal (1): stacked into one row, newest on top, tagged Latest, Similar or Identical. Powered by dupey.

⌨️ Space to see anything

Markdown, code, PDF, video, audio, Office and HWP/HWPX, previewed instantly in Quick Look. HWP pages are drawn locally by rhwp; Hancom Office is not required.

🪟 A real Finder

Tabs and split panes: drag any tab to an edge to split, and every pane keeps its own tabs, so a folder and an AI chat can sit side by side. Folders update live as other apps change them.

And more: hybrid keyword + meaning search reordered by a reranker · scheduled re-indexing that runs even while the app is closed · Ask Dazzi in the Finder and Explorer right-click menu · a dazzi <folder> command for the terminal, installed with the app · light and dark mode · Korean, English, Japanese and Simplified Chinese.

Every place, one search

Connect the apps you already talk in; Dazzi searches them alongside your files.

Private by design

• Embedded on this Mac. Document understanding runs on a local Qwen/Qwen3-Embedding-0.6B model by default. File contents are sent to a remote embedder only if you choose Dazzi cloud embedding (Pro plan) and agree to it in the app.
• Keys stay encrypted. Sign-in tokens and API keys live in the system's encrypted store, never in a config file.
• Bring your own model. Use Dazzi's managed models (Free, Air, Pro, Max), or connect your own provider key and pay them directly.

Develop

[!NOTE] The librarian is not vendored. Lifecycle scripts clone the public AutoRAG repository into the ignored .upstream/autorag at the pinned ref (a release tag, or a commit SHA while the librarian has no tag that carries MinSync 0.4.10), build it, and print the exact revision the app runs against.

Prerequisites: Bun, Node.js 24+, and dupey (cargo install dupey --locked).

make setup   # check toolchain, install dependencies (safe to re-run)
make run     # launch the dev app on free ports, debuggers enabled

Command	What it does
make setup	Verifies bun, Node 24+ and dupey, then installs dependencies
make run	Starts the dev app; prints the renderer URL, main-process inspector, and renderer debug endpoint
make onboarding	Walks first-run onboarding in a disposable sandbox profile
bun run test	Vitest unit tests
bun run typecheck	TypeScript checks
bun run build	Production build into out/
bun run qa	Manual-QA lanes in a throwaway macOS VM (see AGENTS.md)

make run also works from a fresh clone: it installs dependencies when node_modules is missing. Use PORT=9234 bun run dev to pick the Vite port, and AUTORAG_SKIP_DUPEY_CHECK=1 only for an intentional run without version grouping.

Project layout

src/main/        Electron main process: windows, menus, IPC, thumbnails
src/fs-worker/   "Dazzi File System" utility process: listing, watch, dupey
src/preload/     Typed renderer bridge (MessagePort to the fs worker)
src/renderer/    React Finder UI, Tailwind 4 + design tokens
src/preview/     Quick Look panel
src/shared/      Process-boundary contracts and i18n dictionaries
scripts/         Dev tooling, manual-QA lanes, release staging, QA VM
test/            Vitest unit tests

Docs

Document	Covers
docs/architecture.md	Settings, fs worker, onboarding, live sync, attachments, OS menus, QA lanes
DESIGN.md	Styling contract, tokens, components
docs/release.md	Installers, bundled tools, signing
docs/remote-access.md	Browser remote access over LAN, Tailscale, SSH tunnels, and Cloudflare Tunnel
docs/finder-performance.md	Listing and rendering performance
AGENTS.md	Contributor and agent rules, i18n, QA VM

Releases

Publishing a GitHub release builds Dazzi-<version>-arm64.dmg and Dazzi-Setup-<version>-x64.exe. Each installer carries the app, every program it drives, and the local embedding model, so one download is the whole install.

License

Dazzi is Fair Source software released under FSL-1.1-MIT (Functional Source License, Version 1.1, MIT Future License): free to use, read, modify, and redistribute for any non-competing purpose, and each release converts automatically to the MIT License two years after publication. It is built on the open-source @autorag/librarian engine (Apache-2.0). Bundled third-party components and their licenses are listed in THIRD_PARTY_NOTICES.md.