Permanent file hosting for agents, built to showcase ViteHub primitives.

Website · ViteHub · ViteHub docs · ViteHub on GitHub

How it works

1. Blob stores the original file immediately at its Drop key.
2. PNG, JPEG, and WebP uploads continue through image optimization; other files are complete as soon as they are stored.
3. On Cloudflare and Vercel, Queue dispatches image optimization to a Sandbox while the original URL is already usable.
4. Other presets optimize images inline with the same provider-neutral application route.
5. Sharp applies EXIF orientation, strips metadata, and resizes images to fit within 2048 × 2048 without upscaling.
6. Drop replaces the Blob only when the optimized image is smaller. Nitro renders the current stored-file count into the landing page.

The Sandbox is a real npm project in server/sandboxes/image-optimizer. ViteHub materializes that project through Workspace and executes it through the selected Box provider.

Use Drop

Install the public agent skill:

npx skills add https://drop.vitehub.dev

The bundled command uploads and immediately prints the file URL:

node "<skill-directory>/scripts/upload-image.mjs" "/absolute/path/to/file.pdf"

Or use the API directly:

curl --fail-with-body https://drop.vitehub.dev/api/files \
  -F "file=@/absolute/path/to/file.pdf"

Files up to 4 MiB are accepted. PNG, JPEG, and WebP files are optimized when that makes them smaller; PDFs, spreadsheets, documents, archives, and other files are stored unchanged. The public deployment limits uploads to five attempts per source address per minute.

Run locally

You need Node.js 24 or newer and pnpm. The Node preset is the smallest local path and does not require Docker:

pnpm install
VITEHUB_PRESET=node VITEHUB_DEPLOYMENT_NAME=drop-local pnpm build
node .output/server/index.mjs

The landing page is a Nitro-rendered Rendu template because its upload count comes from Blob at request time.

Host it yourself

Set VITEHUB_PRESET to one of ViteHub's built-in presets and VITEHUB_DEPLOYMENT_NAME to namespace provider resources:

VITEHUB_PRESET	Blob storage	Optimization	Rate limit
cloudflare	R2, durable	Queue and Sandbox	Cloudflare binding
vercel	Vercel Blob, durable	Queue and Sandbox	Disabled: no distributed driver
netlify	Netlify Blobs, durable	Inline	Disabled: no distributed driver
node	Filesystem, single host	Inline	Process-local memory
deno	Filesystem compatibility mode, non-durable	Inline	Disabled: no distributed driver

Drop does not substitute process-local rate limiting on horizontally scaled hosts, and it does not claim Queue or Sandbox delivery where the preset cannot provide those guarantees. The generated deployment.json records the selected host, artifact, adapters, guarantees, and explicit unsupported reasons.

The Node artifact needs a persistent volume to retain URLs across process replacement. Deno Deploy has no durable ViteHub Blob adapter, so Drop explicitly selects filesystem compatibility mode; uploads can disappear when a revision or instance is replaced and that target does not satisfy Drop's permanent-storage promise.

Build any target with one preset selection:

VITEHUB_PRESET=netlify VITEHUB_DEPLOYMENT_NAME=my-drop pnpm build

For Deno Deploy, ViteHub stages Sharp's Linux x64 and ARM64 native packages and generates a non-interactive create-or-update runner:

VITEHUB_PRESET=deno VITEHUB_DEPLOYMENT_NAME=my-drop pnpm build
DENO_DEPLOY_ORG=my-org node .output/deploy.mjs

Set DENO_DEPLOY_APP when the app name should differ from VITEHUB_DEPLOYMENT_NAME.

Cloudflare Sandbox requires a Workers Paid plan. After building the cloudflare preset, create the R2 bucket and Queue named by .output/server/wrangler.json, then deploy that generated configuration:

pnpm exec wrangler r2 bucket create my-drop-images
pnpm exec wrangler queues create QUEUE_NAME_FROM_WRANGLER_JSON
pnpm exec wrangler deploy --config .output/server/wrangler.json

ViteHub composes the R2, Queue, Rate Limit, Sandbox, Container, Durable Object, and migration bindings.

Run the complete flow against any deployed origin, declaring whether that preset provides rate limiting:

DROP_ORIGIN=https://YOUR_DOMAIN \
DROP_EXPECT_RATE_LIMIT=1 \
pnpm test:e2e:deployed

Set DROP_EXPECT_RATE_LIMIT=0 for Netlify, Vercel, and Deno. The test drives the public skill command for an image and a document, confirms the document stays unchanged, waits for the optimized image, checks orientation and metadata removal, confirms Blob listing, and verifies the declared rate-limit policy. ViteHub's generated deployment.json records whether the build uses isolated or inline optimization.