Vite + React 19 + @hanzo/gui + @hanzo/iam + @hanzo/base, forked from the canonical hanzo-starter. One org-scoped Base collection (items) behind four @hanzo/gui surfaces: a dense sortable inventory table (amber Low rows, red Out), item detail with stock adjust + reorder point, low-stock alerts, and a manage writer. tsc + vite build green.
5.7 KiB
Stockroom
A dense inventory tracker with low-stock alerts — a real, buildable Hanzo app you fork on hanzo.app and deploy live on Hanzo Cloud.
Track every SKU by quantity and location, watch on-hand counts against reorder points, and let low stock flag itself in amber before it becomes a stockout.
- UI —
@hanzo/gui(the Hanzo design system) under Vite + React 19. 100% gui primitives — no Tailwind, no second kit. - Auth —
@hanzo/iam, OAuth2 PKCE against hanzo.id. No local passwords — IAM owns every credential interaction. - Data —
@hanzo/base, the IAM-native, org-scoped data plane. Theitemsinventory is a real Base collection.
The app
Four surfaces over one items collection:
- Inventory — a dense, sortable, searchable data-table of every SKU (SKU, item, location, on-hand, reorder point, status). Rows flag themselves: amber when on-hand is at/under the reorder point, red at zero.
- Item detail — status, on-hand vs reorder point and a suggested order size, plus the writers: receive/pick units, change the reorder point, delete.
- Alerts — every item at or below its reorder point (out first, then the deepest shortfall), each with a suggested order quantity.
- Manage — add a SKU with its starting count, reorder point and location.
Stack (pinned)
| Package | Version |
|---|---|
react / react-dom |
^19.2.4 |
@hanzo/gui + @hanzogui/config |
7.3.0 |
@hanzo/iam |
^0.13.1 |
@hanzo/base |
^0.2.1 |
vite |
^6 (@vitejs/plugin-react) |
typescript |
5.9.3 |
Run it
npm install
npm run dev # http://localhost:5173
npm run build # tsc --noEmit && vite build -> dist/
npm run preview # serve the production build (SPA fallback on)
Out of the box it runs against live Hanzo (hanzo.id + api.hanzo.ai) — no
config needed to see the sign-in flow. Copy .env.example to .env to point at
a different environment.
Environment contract
Only VITE_-prefixed vars reach the browser (this is a static SPA — there is no
server). Defaults in parentheses.
| Var | Purpose |
|---|---|
VITE_HANZO_IAM_URL (https://hanzo.id) |
OIDC issuer. |
VITE_HANZO_CLIENT_ID (hanzo-app) |
IAM application (<org>-<app>). Its redirect-URI list must allow this deploy's /auth/callback — see Ambient IAM. |
VITE_HANZO_REDIRECT_URI (${origin}/auth/callback) |
PKCE redirect. |
VITE_HANZO_BASE_URL (https://api.hanzo.ai) |
Browser-reachable Hanzo Base data plane. Deploy injects the provisioned URL. |
VITE_HANZO_API_URL / VITE_HANZO_API_KEY |
Optional — only if the app calls the Hanzo LLM gateway. Unused by Stockroom; never commit a real key. |
How auth works — ambient IAM
login() starts an OAuth2 PKCE S256 redirect to hanzo.id; hanzo.id returns
to /auth/callback, where handleCallback() exchanges the code for tokens
(stored in localStorage, refresh-aware via offline_access). Every deployed
app is a static site at <slug>.hanzo.app; there is no server token — the
SPA authenticates the user in the browser and carries the resulting IAM JWT to
Base. "Ambient" means the app just reads the signed-in user via that token.
The one deploy requirement: the IAM client (VITE_HANZO_CLIENT_ID) must list
this origin's /auth/callback as an allowed redirect URI. Register a
https://*.hanzo.app/auth/callback wildcard on the shared client so every
forked app works, or register a dedicated hanzo-<app> client per template.
How data works — Base from schema.sql
schema.sql is the app's databaseSchema (SQL DDL). On publish,
Hanzo Cloud translates the items CREATE TABLE into a Hanzo Base collection
(provisionBaseFromDDL, additive + idempotent). Base manages
id/created/updated/owner/org, stamps owner+org from the verified
IAM principal, and scopes every row to the caller's org via the rule
@request.auth.org_id = org — a teammate in your org sees the row; other orgs
cannot. At runtime the views read/write that collection through
@hanzo/base/react (useQuery/useMutation) carrying the IAM token. Keep
schema.sql in lockstep with src/lib/stock.ts and the views.
Deploy — Hanzo Cloud
hanzo.yml declares a static build (npm run build → dist/,
served at <slug>.hanzo.app) plus the Base schema to provision and the env to
inject. Do not build a container image locally — Hanzo Cloud owns builds and
deploys. CI here only proves the template compiles green.
Layout
src/
main.tsx entry
providers.tsx GuiProvider -> IamProvider -> BaseProvider(client=IAM-token)
app.tsx route (/auth/callback) + auth gate
gui.config.ts createGui(defaultConfig from @hanzogui/config/v5)
iam.config.ts IAM PKCE config
env.ts the VITE_ env contract, one place
lib/base.ts BaseClient carrying the IAM bearer token
lib/stock.ts Item type, status/severity/reorder math, the warehouse palette
auth/callback.tsx PKCE return leg
views/
signed-out.tsx landing + illustrative preview table
home.tsx shell: brand bar + tabs + KPI strip + detail leg
inventory.tsx the dense, sortable stock table
detail.tsx item detail + stock adjust / reorder point / delete
alerts.tsx low-stock alerts, most urgent first
manage.tsx add a SKU
brand.tsx the Stockroom mark
schema.sql databaseSchema -> the `items` Base collection on publish
hanzo.yml Hanzo Cloud build/deploy manifest