netlify-blobs

Netlify Blobs

Netlify Blobs is zero-config object storage for files and assets: images, documents, uploads, exports, cached binary artifacts. Available from any Netlify compute (functions, edge functions, framework server routes). No provisioning required.

Not for dynamic data. If the project needs to store records, user data, application state, or anything queryable, use Netlify Database instead — see netlify-database/SKILL.md. Reach for Blobs when the thing you're storing is a file or an asset blob, not a record.

npm install @netlify/blobs

Before you build

If the prompt didn't already specify, ask the user a few short questions before scaffolding any blob storage — answers shape access patterns, scoping, and how the assets are served back to clients:

• What kind of asset? (User uploads, exported documents, cached binaries, generated images — drives the storage and serving pattern.)
• Who should be able to read it? Public (anyone with a URL, or an unauthenticated endpoint that streams the blob) or private (only authenticated users, gated by your server code)? Blobs have no built-in access control — the serving layer is the gate. When in doubt, default to private; making something public later is easy, while pulling back data that was inadvertently exposed is not.
• Site-scoped or deploy-scoped? Site-scoped (getStore()) persists across deploys — the right default for user data. Deploy-scoped (getDeployStore()) is tied to a single deploy and disappears when that deploy is replaced — use only when the lifecycle should match a deploy (e.g., per-deploy build artifacts).
• Roughly how big and how many? Helps choose between a single large blob vs many small keyed blobs, and informs whether you'll need list({ prefix: ... }) patterns.

If you don't have preferences here, tell me what the assets are and I'll pick sensible defaults — typically site-scoped with private access, served through an authenticated function.

Getting a Store

import { getStore } from "@netlify/blobs";

const store = getStore({ name: "my-store" });

// Use "strong" consistency when you need immediate reads after writes
const store = getStore({ name: "my-store", consistency: "strong" });

CRUD Operations

These are the only store methods. Do not invent others.

Create / Update

// String or binary data
await store.set("key", "value");
await store.set("key", fileBuffer);

// With metadata
await store.set("key", data, {
  metadata: { contentType: "image/png", uploadedAt: new Date().toISOString() },
});

// JSON data
await store.setJSON("key", { name: "Example", count: 42 });

Read

// Text (default)
const text = await store.get("key");                    // string | null

// Typed retrieval
const json = await store.get("key", { type: "json" });  // object | null
const stream = await store.get("key", { type: "stream" });
const blob = await store.get("key", { type: "blob" });
const buffer = await store.get("key", { type: "arrayBuffer" });

// With metadata
const result = await store.getWithMetadata("key");
// { data: any, etag: string, metadata: object } | null

// Metadata only (no data download)
const meta = await store.getMetadata("key");
// { etag: string, metadata: object } | null

Delete

await store.delete("key");

List

const { blobs } = await store.list();
// blobs: [{ etag: string, key: string }, ...]

// Filter by prefix
const { blobs } = await store.list({ prefix: "uploads/" });

store.list() auto-paginates: a plain await store.list() transparently fetches every page and returns the complete blobs array — you do NOT hand-roll page cursors or offsets. For a very large store, pass { paginate: true } to get an async iterator and stream results a page at a time instead of buffering every key in memory:

for await (const page of store.list({ paginate: true })) {
  for (const { key } of page.blobs) {
    // handle each key
  }
}

Pass { directories: true } to group keys by the / delimiter (folder-style): the result's blobs holds keys at the current level and directories holds the common prefixes, which you drill into with prefix. Keys are a flat namespace — / is only a naming convention that prefix and directories let you navigate.

Store Types

• Site-scoped (getStore()): Persist across all deploys. Use for most cases.
• Deploy-scoped (getDeployStore()): Tied to a specific deploy lifecycle.

A site-scoped store is shared across ALL deploy contexts. Production, deploy previews, and branch deploys all read and write the same getStore() store — unlike Netlify Database, which forks a separate branch per preview, Blobs does not isolate previews. Code running on a deploy preview reads, overwrites, and deletes the same production data. Don't run destructive tests or seed throwaway data against a getStore() store from a preview — it hits production. When you need per-context isolation, use getDeployStore(), or partition by deploy context with a context-specific store name or key prefix.

Consistency and concurrency

Blobs are eventually consistent by default: an immediate read right after a write may return the previous value or null. Opt into strong consistency when you need read-your-writes. You can set it once on the store, or request it per read:

import { getStore } from "@netlify/blobs";

const store = getStore({ name: "my-store", consistency: "strong" });

// or just for a single read that must see the latest write:
const fresh = await store.get("key", { consistency: "strong" });

Strong reads are slower than eventual reads, so don't make everything strong "to be safe" — reserve it for the reads that genuinely need the latest write (typically a read right after a write in the same request). For read-heavy access to data that rarely changes, the default eventual consistency is faster and is the right choice.

Blobs has no concurrency control: there is no locking and there are no transactions, and concurrent writes to the same key are last-write-wins — one silently overwrites the other. Do NOT build counters, balances, or any read-modify-write logic over a single blob key and expect it to be correct under concurrent traffic (two requests can both read the old value and both write back, losing an update). When you need atomic or transactional updates, use Netlify Database (see netlify-database/SKILL.md), which provides real transactions — not Blobs.

Limits

Limit	Value
Max object size	5 GB
Metadata per object	2 KB
Store name max length	64 bytes
Key max length	600 bytes

Object metadata is capped at 2 KB per object — it's for small descriptors (content type, size, timestamps, a status flag), not a place to stash large JSON. Anything bigger belongs in the blob value itself, not in metadata.

Local Development

Local dev uses a sandboxed store (separate from production). For Vite-based projects, install @netlify/vite-plugin to enable local Blobs access. Otherwise, use netlify dev.

Common error: "The environment has not been configured to use Netlify Blobs" — install @netlify/vite-plugin or run via netlify dev.

Inspecting blobs from the CLI

The Netlify CLI can read and write blobs directly — useful for debugging, seeding, or a one-off fix without writing and deploying a function:

netlify blobs:list <store-name>
netlify blobs:get <store-name> <key>
netlify blobs:set <store-name> <key> <value>
netlify blobs:delete <store-name> <key>

These act on the linked site's store, so link the project first (netlify link). Reach for these documented subcommands for manual inspection or repair rather than the raw API.

Uploading blobs at build time

You don't have to write blobs from runtime code — you can seed a store during the build by writing files into a special directory. Files placed in .netlify/blobs/deploy/ during the build are uploaded to a deploy-scoped store and are then readable at runtime via getDeployStore(). The path under that directory becomes the blob key (so .netlify/blobs/deploy/products/1.json is stored under the key products/1.json). This avoids a runtime function looping over store.set on a cold start.

To attach metadata to a build-time blob, add a JSON sidecar whose name is the blob's filename prefixed with $ and suffixed with .json — metadata for logo.png goes in $logo.png.json. Read these blobs back with getDeployStore(), not getStore(): they live in the deploy-scoped store and are replaced when the deploy is replaced.

When a store operation fails

If a get/set call throws in a deployed function, don't guess at a fix or route around it — the exact error is in the function logs, and it almost always names the cause. Read it first. Common causes: the store isn't reachable from the calling context, a missing or mismatched store name, or a read-after-write timing gap (an immediate read of a just-written key — use consistency: "strong" when you need read-your-writes).

The store exposes only the documented methods above; there is no lower-level REST endpoint to fall back on. If the logs don't resolve it, report the exact error plus the affected site/deploy to the user and stop. Reaching around a failing store — direct https://api.netlify.com/... calls, reading auth tokens off disk, or inventing endpoints — can't work (those aren't supported surfaces) and risks corrupting or losing the very data you're trying to save.