Strut

An HTML5 GUI authoring tool for spatial presentations — build a deck of slides, place rich content on each, arrange the slides in 3-D space, and play the deck as a camera flight through that world (the impress.js model, made visual and editable).

Stack

• React 19 + TanStack Start (file-based routing, SSR) on Vite.
• Rindle for the data layer (wired): SQL migrations as source of truth → generated TypeScript schema → optimistic local store + live windowed queries + named mutators, with a rindled daemon behind the app's own server routes. See RINDLE_NOTES.md for the write-up.
• Plain CSS for the editor chrome (src/strut.css); Tailwind is available for one-offs.

Architecture

• Rindle fleet — a rindle-replicator write-master + one rindled follower (which owns the live-query WebSocket), fronted by a rindle-dev-edge ingress that serves control, reads and ws on a single port: 127.0.0.1:22050. Shape comes from rindle.ncl; rindle up/dev render and supervise it.
• API — TanStack Start server routes (src/routes/api.rindle.*) host the stateless Rindle API (server/rindle-api.ts): they validate args, run authoritative SQL mutators, and register the named queries. Same-origin, no separate process. Image uploads (server/upload.ts) go to Cloudflare R2 — a native bucket binding on Workers, the S3 API on other hosts, else a local dev fallback. Mirrors the predicted client mutators in shared/app-def.ts.
• Browser client (src/rindle/*) — the optimistic store (@rindle/optimistic + WASM), useQuery live reads, and app.mutate.* writes, posting to /api/rindle/*. The live-query WebSocket connects directly to the fleet ingress (:22050).

Schema lives in migrations/; shared/ holds the generated schema, query builder, named queries, and client mutators (imported by both browser and server). App code is in src/ (routes/, editor/, rindle/).

Getting started

The normal local run path is a single command:

pnpm install

pnpm dev

Then open http://localhost:3000.

pnpm dev is rindle dev … -- pnpm dev:web: the CLI owns the fleet for the lifetime of the web command, then hands it the connection bindings as env vars.

• rindle dev --migrate --gen shared/schema.ts --watch — renders rindle.ncl, supervises the fleet, applies migrations, regenerates shared/schema.ts, and keeps watching migrations/.
• vite dev --port 3000 — starts the TanStack Start app on http://localhost:3000. The Rindle API and image upload endpoints are served by this same web process under /api/rindle/*; there is no separate API server to start.

Local state lives in master.db (the replicator's write-master), follower-0.db, and .uploads/. Image uploads work with no config by using the local fallback; copy .env.example to .env only if you want uploads stored in Cloudflare R2. vite.config.ts loads .env for server-side values during vite dev.

If you need to run the processes separately:

pnpm daemon   # fleet + migration/schema watcher
pnpm dev:web  # web app + same-origin API routes; expects the fleet to already be running

Both legs default to the fleet ingress — http://127.0.0.1:22050 for control/reads and ws://127.0.0.1:22050 for the live-query WebSocket. rindle dev/exec inject the real bindings (RINDLE_URL, and under exec also RINDLE_DAEMON_URL/RINDLE_FLEET_WS), so the defaults only matter when you run pnpm dev:web on its own. Override with RINDLE_DAEMON_URL for the server/API side, RINDLE_DAEMON_WS for the WS URL the server hands the browser (/api/rindle/config), or VITE_RINDLE_WS as a build-time client fallback. See server/rindleEnv.ts.

Other scripts:

pnpm build            # production build (client + SSR + API routes); does not start the daemon
pnpm preview          # preview the built app; expects a reachable daemon
pnpm generate-routes  # regenerate src/routeTree.gen.ts
pnpm setup            # one-shot migrate + schema regen against a running daemon
pnpm test             # vitest
pnpm lint             # eslint
pnpm check            # prettier check

Deploying to Cloudflare

The web app (SSR + /api/rindle/* routes) deploys to Cloudflare Workers, with image uploads in R2 via a native bucket binding. The rindled daemon can't run on Workers and must be hosted separately. pnpm deploy builds the Worker (CF=1 vite build) and ships it with wrangler. Local pnpm dev/pnpm build stay on Node and are unaffected. See docs/DEPLOY_CLOUDFLARE.md for the full guide (daemon hosting, R2 setup, secrets, deploy steps).

AI: who pays

The ✨ features (Chat and everything it can do — author slides, arrange, restyle, edit) run on a model the user supplies: connect an OpenRouter key once and AI turns on across the editor, on their credits, with no account required.

That is the only inference path. Without a connected key, every ✨ route answers 402 and the editor shows a connect-a-model gate. Strut has no deployment-funded provider keys, AI quota ledger, generated- image backend, or speech-to-text backend to monitor. Image uploads and openly licensed image search remain.

Commercial / hosted (optional)

Strut is fully open source and self-hostable for free. The official hosted Strut (marketing page + paid Pro accounts via Stripe) is a private overlay kept out of this repo and merged in only at deploy time — so a clone has no billing, no marketing, and no paywall, and behaves exactly as it does today. This follows the same opt-in posture as the analytics above.

The repo ships the open-core seam (#commercial) that an overlay plugs into: the app reads an entitlement (server/entitlements.ts) for hosted storage, private-deck, and white-label features, and renders an optional Upgrade affordance — all inert without an overlay. pnpm deploy builds the free app on a single host; pnpm deploy:pro builds the app plus the overlay as one Worker (marketing on strut.io, the app on strut.io/app). See docs/COMMERCIAL_OVERLAY.md.

Analytics

Strut ships with optional, privacy-first product analytics via Umami — cookieless, no personal data, no consent banner. It records only which features get used (slides generated, AI arrange applied, present started, exports, sign-ins) — never your deck content, prompts, or any PII.

It is off by default and off in every clone. Nothing is collected and no script loads unless the build sets VITE_UMAMI_SRC (+ VITE_UMAMI_ID) — so cloning this repo phones home to no one. To enable it on your own deployment, set those two build-time vars (see .env.example); point VITE_UMAMI_SRC at Umami Cloud or your own self-hosted instance. Implementation lives in src/lib/analytics.ts.