Getting started

Three ways in: desktop installer (least setup), from source (if you want to change code), or desktop dev mode (when working on the Rust shell or sidecar).

1. Desktop installer

Channel	Windows artifact	Runtime and size
`fat`	`Hetu_x.y.z_x64-setup.exe` / `.msi`	Bundles the .NET 10 runtime, ready after install, ~100 MB+
`slim`	`Hetu.Slim._x.y.z_x64-setup.exe` / `.msi`	Requires the .NET 10 runtime, smaller download
Linux	`.AppImage` / `.deb`	AppImage needs `chmod +x`

The two channels are independent update streams: a slim install only receives latest-slim.json and never gets overwritten by the fat build.

Closing the window minimizes to the tray by default (the backend keeps running, background jobs continue). Quit from the tray menu. This can be turned off in Settings → App.

2. From source

git clone https://github.com/wosledon/Hetu
cd Hetu

pwsh ./scripts/start.ps1        # Windows: starts backend + frontend
./scripts/start.sh              # Linux / macOS

# or separately
dotnet run --project src/Hetu.Api --urls "http://localhost:5000"
cd frontend && npm install && npm run dev   # http://localhost:5174

Address	Purpose
`http://localhost:5174`	Frontend UI
`http://localhost:5000/api`	Backend API
`http://localhost:5000/scalar/v1`	API reference (Scalar)
`http://localhost:5000/api/health`	Health check used by the desktop shell

3. Desktop dev mode

Runs dotnet + vite + tauri in parallel — useful when changing the Rust shell, the tray or the updater:

pwsh ./scripts/desktop-dev.ps1

If port 5000 already has a backend running, the shell reuses it instead of starting another one.

First configuration

Step	Where	Notes
Add a provider	Models → add provider	Protocol: OpenAI-compatible or Anthropic; set Base URL and API key (local inference servers work too)
Add models	Under that provider	Purpose: `chat` / `embedding` / `completion`
Set defaults	Settings → Default models	Per scenario: chat, code, Wiki, graph, organize, note AI, query rewrite
Feed knowledge	Knowledge base	Add URLs or upload files; for notes use “Generate index” in the editor toolbar
Try it	Code page	Toggle “Knowledge base” above the input to inject retrieval results

Anthropic has no public embedding API — use an OpenAI-compatible provider for the embedding purpose. Vector dimensions must match the model (Embedding:Dimensions, default 1536).

Where data lives

How you run it	Location
Desktop (Windows)	`%LOCALAPPDATA%\Hetu`: `hetu.db` (+ WAL) and `logs/`
From source	Whatever `HETU_DATA_DIR` points at; otherwise `hetu.db` in the working directory

Let it change code for you: Code sessions.
Turn documents into searchable knowledge: Knowledge base, Graph & Wiki.
Keep cost visible: Models & usage.