Chat

Desktop AI client — streaming SSE to any OpenAI-compatible gateway, threads and API key kept out of web storage.

A Hanzo desktop template: one codebase, three installers (macOS, Windows, Linux) and a web build. Fork it, npm i, ship.

desktop-chat

Why this exists

A native app cannot be iframed, so the gallery preview is a real build of the same UI — not a screenshot and not a dead link. src/native.ts is the only file that knows whether it is running inside the Tauri window or a browser tab; everything above it is one app.

Stack

Layer Choice
Shell Tauri v2 (Rust core, system webview — installers are single-digit MB, not 150 MB)
UI React 19 + Vite 7 + TypeScript
Backend @hanzo/base — collections, realtime, CRDT
Runtime a.hanzo.ai analytics + chat, loaded from index.html
Native API key + thread history in the OS config dir (0600), never in localStorage

Run

npm install
npm run desktop        # Tauri dev window
npm run dev            # browser, same UI, web fallbacks
npm run desktop:build  # installers for the host OS
npm run build          # static web build -> dist/

Linux build deps: libwebkit2gtk-4.1-dev libgtk-3-dev librsvg2-dev patchelf.

Backend

src/base.ts points at https://base.hanzo.ai. Point a fork at its own with VITE_BASE_URL. Base is optional — the app renders and works offline, the status bar just says so.

Release

Push a v* tag. .github/workflows/release.yml fans out to macOS (arm64 + x86_64), Windows and Linux runners and attaches .dmg / .msi / .AppImage / .deb to the GitHub release. Desktop bundles must be built on their target OS, which is why this one workflow uses per-platform runners.

Upstream & license

Scaffold derived from Tauri v2 app template + OpenAI-compatible streaming client (https://github.com/tauri-apps/tauri/tree/dev/examples/api), MIT / Apache-2.0. This template is MIT — see LICENSE. Every dependency is MIT/Apache-2.0/BSD; nothing copyleft is pulled in, so a fork can be commercialised without contaminating it.

S
Description
Desktop AI client — streaming SSE to any OpenAI-compatible gateway, threads and API key kept out of web storage.
Readme MIT
213 KiB
Languages
TypeScript 62.5%
CSS 25%
Rust 9.6%
HTML 2.9%