feathersjs · daffl · Dec 20, 2025 · Dec 19, 2025 · Dec 19, 2025 · Dec 19, 2025
diff --git a/docs/api/client/rest.md b/docs/api/client/rest.md
@@ -22,7 +22,7 @@ The following chapter describes the use of
 npm install @feathersjs/rest-client --save
 ```
 
-`@feathersjs/rest-client` allows to connect to a service exposed through a REST HTTP transport (e.g. with [Koa](../koa.md#rest) or [Express](../express.md#rest)) using [fetch](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API), [Superagent](https://github.com/ladjs/superagent) or [Axios](https://github.com/mzabriskie/axios).
+`@feathersjs/rest-client` allows to connect to a service exposed through a REST HTTP transport (e.g. with [Koa](../koa.md#rest) or [Express](../express.md#rest)) using [fetch](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API), [Superagent](https://github.com/ladjs/superagent) or [Axios](https://github.com/mzabriskie/axios).
 
 <BlockQuote type="info">
 
@@ -226,6 +226,91 @@ File uploads use the native `Request.formData()` API which buffers the entire re
 
 </BlockQuote>
 
+### Streaming Uploads
+
+The REST client supports streaming data to services using `ReadableStream`. This is useful for large file uploads, real-time data ingestion, or piping data directly to storage without buffering.
+
+```ts
+// Stream a file to a service
+const file = fileInput.files[0]
+const stream = file.stream()
+
+const result = await app.service('uploads').create(stream, {
+  headers: {
+    'Content-Type': file.type,
+    'X-Filename': file.name
+  }
+})
+```
+
+On the server, the service receives the `ReadableStream` directly:
+
+```ts
+class UploadService {
+  async create(stream: ReadableStream, params: Params) {
+    const filename = params.headers['x-filename']
+    const contentType = params.headers['content-type']
+
+    // Pipe directly to storage - no buffering
+    await storage.upload(filename, stream, { contentType })
+
+    return { filename, uploaded: true }
+  }
+}
+```
+
+The stream can be piped directly to cloud storage (S3, R2, etc.) without loading the entire file into memory:
+
+```ts
+async create(stream: ReadableStream, params: Params) {
+  // Stream directly to R2/S3
+  await env.MY_BUCKET.put(params.headers['x-filename'], stream)
+  return { success: true }
+}
+```
+
+For more complex metadata, you can stringify an object into a header:
+
+```ts
+// Client
+const file = fileInput.files[0]
+
+await app.service('csv-import').create(file.stream(), {
+  headers: {
+    'Content-Type': 'text/csv',
+    'X-Import-Options': JSON.stringify({
+      filename: file.name,
+      tableName: 'products',
+      skipHeader: true
+    })
+  }
+})
+
+// Server
+async create(stream: ReadableStream, params: Params) {
+  const options = JSON.parse(params.headers['x-import-options'])
+  // options.filename, options.tableName, options.skipHeader
+}
+```
+
+<BlockQuote type="warning" label="Header size limits">
+
+HTTP headers are typically limited to 8KB total. Keep metadata small - use headers for filenames, options, and IDs, not large data payloads.
+
+</BlockQuote>
+
+<BlockQuote type="info" label="Content-Type">
+
+If no `Content-Type` header is specified, streaming requests default to `application/octet-stream`. Any content type not recognized as JSON, form-urlencoded, or multipart will be streamed through to the service.
+
+</BlockQuote>
+
+<BlockQuote type="warning" label="REST only">
+
+Streaming uploads are only supported with the REST/HTTP transport. Socket.io does not support streaming request bodies.
+
+</BlockQuote>
+
 ### Custom Methods
 
 On the client, [custom service methods](../services.md#custom-methods) registered using the `methods` option when registering the service via `restClient.service()`:

diff --git a/docs/api/hooks.md b/docs/api/hooks.md
@@ -199,6 +199,74 @@ If you want to inspect the hook context, e.g. via `console.log`, the object retu
 
 </BlockQuote>
 
+#### Working with Streams
+
+When using [streaming uploads](./client/rest.md#streaming-uploads), `context.data` will be a `ReadableStream`. Since streams can only be consumed once, around hooks are the recommended way to work with streaming data. Here are common patterns:
+
+**Passing streams through unchanged:**
+
+If you only need to validate metadata or check permissions, you can let the stream pass through to the service:
+
+```ts
+app.service('uploads').hooks({
+  around: {
+    create: [
+      async (context: HookContext, next: NextFunction) => {
+        // Validate using headers - don't consume the stream
+        const contentType = context.params.headers?.['content-type']
+        if (!contentType?.startsWith('image/')) {
+          throw new BadRequest('Only images are allowed')
+        }
+
+        // Stream passes through unchanged
+        await next()
+      }
+    ]
+  }
+})
+```
+
+**Wrapping streams with transforms:**
+
+You can wrap the incoming stream with a transform stream for processing:
+
+```ts
+import { TransformStream } from 'node:stream/web'
+
+app.service('uploads').hooks({
+  around: {
+    create: [
+      async (context: HookContext, next: NextFunction) => {
+        const originalStream = context.data as ReadableStream
+
+        // Create a transform that tracks bytes
+        let totalBytes = 0
+        const countingTransform = new TransformStream({
+          transform(chunk, controller) {
+            totalBytes += chunk.length
+            controller.enqueue(chunk)
+          }
+        })
+
+        // Replace with transformed stream
+        context.data = originalStream.pipeThrough(countingTransform)
+
+        await next()
+
+        // After service completes, totalBytes is available
+        context.result.size = totalBytes
+      }
+    ]
+  }
+})
+```
+
+<BlockQuote type="warning" label="Important">
+
+Streams can only be consumed once. If you need to read the stream content in a hook (e.g., for validation), you must either buffer the entire stream or use a tee/transform approach. For large files, prefer validating metadata from headers rather than consuming the stream.
+
+</BlockQuote>
+
 ### `context.error`
 
 `context.error` is a **writeable** property with the error object that was thrown in a failed method call. It can be modified to change the error that is returned at the end.