Upload folder using huggingface_hub (part 11)
96c8713 verified about 2 months ago
34.9 kB
<br />
<p align="center">
  <a href="https://supabase.io">
        <picture>
      <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/supabase/supabase/master/packages/common/assets/images/supabase-logo-wordmark--dark.svg">
      <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/supabase/supabase/master/packages/common/assets/images/supabase-logo-wordmark--light.svg">
      <img alt="Supabase Logo" width="300" src="https://raw.githubusercontent.com/supabase/supabase/master/packages/common/assets/images/logo-preview.jpg">
    </picture>
  </a>
  <h1 align="center">Supabase Storage JS SDK</h1>

  <h3 align="center">JavaScript SDK to interact with Supabase Storage, including file storage and vector embeddings.</h3>

  <p align="center">
    <a href="https://supabase.com/docs/guides/storage">Guides</a>
    ·
    <a href="https://supabase.com/docs/reference/javascript/storage-createbucket">Reference Docs</a>
    ·
    <a href="https://supabase.github.io/supabase-js/storage-js/v2/spec.json">TypeDoc</a>
  </p>
</p>
<div align="center">

[![Build](https://github.com/supabase/supabase-js/workflows/CI/badge.svg)](https://github.com/supabase/supabase-js/actions?query=branch%3Amaster)
[![Package](https://img.shields.io/npm/v/@supabase/storage-js)](https://www.npmjs.com/package/@supabase/storage-js)
[![License: MIT](https://img.shields.io/npm/l/@supabase/supabase-js)](#license)
[![pkg.pr.new](https://pkg.pr.new/badge/supabase/storage-js)](https://pkg.pr.new/~/supabase/storage-js)

</div>

## Requirements

- **Node.js 22 or later** (Node.js 20 support dropped in v2.110.0)
- For browser support, all modern browsers are supported

> ⚠️ **Node.js 18 Deprecation Notice**
>
> Node.js 18 reached end-of-life on April 30, 2025. As announced in [our deprecation notice](https://github.com/orgs/supabase/discussions/37217), support for Node.js 18 was dropped on October 31, 2025.

> ⚠️ **Node.js 20 Deprecation Notice**
>
> Node.js 20 reached end-of-life on April 30, 2026. As announced in [our deprecation notice](https://github.com/orgs/supabase/discussions/45715), support for Node.js 20 was dropped in v2.110.0.

## Features

- **File Storage**: Upload, download, list, move, and delete files
- **Access Control**: Public and private buckets with fine-grained permissions
- **Signed URLs**: Generate time-limited URLs for secure file access
- **Image Transformations**: On-the-fly image resizing and optimization
- **Vector Embeddings**: Store and query high-dimensional embeddings with similarity search
- **Analytics Buckets**: Iceberg table-based buckets optimized for analytical queries and data processing

## Quick Start Guide

### Installing the module

```bash
npm install @supabase/storage-js
```

### Connecting to the storage backend

There are two ways to use the Storage SDK:

#### Option 1: Via Supabase Client (Recommended)

If you're already using `@supabase/supabase-js`, access storage through the client:

```js
import { createClient } from '@supabase/supabase-js'
// Use publishable key for frontend applications
const supabase = createClient('https://<project_ref>.supabase.co', '<your-publishable-key>')
// Access storage
const storage = supabase.storage
// Access different bucket types
const regularBucket = storage.from('my-bucket')
const vectorBucket = storage.vectors.from('embeddings-bucket')
const analyticsBucket = storage.analytics // Analytics API
```

#### Option 2: Standalone StorageClient

For backend applications or when you need to bypass Row Level Security:

```js
import { StorageClient } from '@supabase/storage-js'
const STORAGE_URL = 'https://<project_ref>.supabase.co/storage/v1'
const SERVICE_KEY = '<your-secret-key>' // Use secret key for backend operations
const storageClient = new StorageClient(STORAGE_URL, {
  apikey: SERVICE_KEY,
  Authorization: `Bearer ${SERVICE_KEY}`,
})
// Access different bucket types
const regularBucket = storageClient.from('my-bucket')
const vectorBucket = storageClient.vectors.from('embeddings-bucket')
const analyticsBucket = storageClient.analytics // Analytics API
```

> **When to use each approach:**
>
> - Use `supabase.storage` when working with other Supabase features (auth, database, etc.) in frontend applications
> - Use `new StorageClient()` for backend applications, Edge Functions, or when you need to bypass RLS policies

> **Note:** Refer to the [Storage Access Control guide](https://supabase.com/docs/guides/storage/access-control) for detailed information on creating RLS policies.

### Understanding Bucket Types

Supabase Storage supports three types of buckets, each optimized for different use cases:

#### 1. Regular Storage Buckets (File Storage)

Standard buckets for storing files, images, videos, and other assets.

```js
// Create regular storage bucket
const { data, error } = await storageClient.createBucket('my-files', {
  public: false,
})
// Upload files
await storageClient.from('my-files').upload('avatar.png', file)
```

**Use cases:** User uploads, media assets, documents, backups

#### 2. Vector Buckets (Embeddings Storage)

Specialized buckets for storing and querying high-dimensional vector embeddings.

```js
// Create vector bucket
await storageClient.vectors.createBucket('embeddings-prod')
// Create index and insert vectors
const bucket = storageClient.vectors.from('embeddings-prod')
await bucket.createIndex({
  indexName: 'documents',
  dimension: 1536,
  distanceMetric: 'cosine',
})
```

**Use cases:** Semantic search, AI-powered recommendations, similarity matching

**[See full Vector Embeddings documentation below](#vector-embeddings)**

#### 3. Analytics Buckets

Specialized buckets using Apache Iceberg table format, optimized for analytical queries and large-scale data processing.

```js
// Create analytics bucket
await storageClient.analytics.createBucket('analytics-data')
// List analytics buckets
const { data, error } = await storageClient.analytics.listBuckets()
// Delete analytics bucket
await storageClient.analytics.deleteBucket('analytics-data')
```

**Use cases:** Time-series data, analytical queries, data lakes, large-scale data processing, business intelligence

**[See full Analytics Buckets documentation below](#analytics-buckets)**

---

### Handling resources

#### Handling Storage Buckets

- Create a new Storage bucket:

  ```js
  const { data, error } = await storageClient.createBucket(
    'test_bucket', // Bucket name (must be unique)
    { public: false } // Bucket options
  )
  ```

- Retrieve the details of an existing Storage bucket:

  ```js
  const { data, error } = await storageClient.getBucket('test_bucket')
  ```

- Update a new Storage bucket:

  ```js
  const { data, error } = await storageClient.updateBucket(
    'test_bucket', // Bucket name
    { public: false } // Bucket options
  )
  ```

- Remove all objects inside a single bucket:

  ```js
  const { data, error } = await storageClient.emptyBucket('test_bucket')
  ```

- Delete an existing bucket (a bucket can't be deleted with existing objects inside it):

  ```js
  const { data, error } = await storageClient.deleteBucket('test_bucket')
  ```

- Retrieve the details of all Storage buckets within an existing project:

  ```js
  // List all buckets
  const { data, error } = await storageClient.listBuckets()
  // List buckets with options (pagination, sorting, search)
  const { data, error } = await storageClient.listBuckets({
    limit: 10,
    offset: 0,
    sortColumn: 'created_at',
    sortOrder: 'desc',
    search: 'prod',
  })
  ```

#### Handling Files

- Upload a file to an existing bucket:

  ```js
  const fileBody = ... // load your file here
  const { data, error } = await storageClient.from('bucket').upload('path/to/file', fileBody)
  ```

  > Note:  
  > The path in `data.Key` is prefixed by the bucket ID and is not the value which should be passed to the `download` method in order to fetch the file.  
  > To fetch the file via the `download` method, use `data.path` and `data.bucketId` as follows:
  >
  > ```javascript
  > const { data, error } = await storageClient.from('bucket').upload('/folder/file.txt', fileBody)
  > // check for errors
  > const { data2, error2 } = await storageClient.from(data.bucketId).download(data.path)
  > ```

  > Note: The `upload` method also accepts a map of optional parameters. For a complete list see the [Supabase API reference](https://supabase.com/docs/reference/javascript/storage-from-upload).

- Download a file from an exisiting bucket:

  ```js
  const { data, error } = await storageClient.from('bucket').download('path/to/file')
  ```

- List all the files within a bucket:

  ```js
  const { data, error } = await storageClient.from('bucket').list('folder')
  ```

  > Note: The `list` method also accepts a map of optional parameters. For a complete list see the [Supabase API reference](https://supabase.com/docs/reference/javascript/storage-from-list).

- Replace an existing file at the specified path with a new one:

  ```js
  const fileBody = ... // load your file here
  const { data, error } = await storageClient
    .from('bucket')
    .update('path/to/file', fileBody)
  ```

  > Note: The `upload` method also accepts a map of optional parameters. For a complete list see the [Supabase API reference](https://supabase.com/docs/reference/javascript/storage-from-upload).

- Move an existing file:

  ```js
  const { data, error } = await storageClient
    .from('bucket')
    .move('old/path/to/file', 'new/path/to/file')
  ```

- Delete files within the same bucket:

  ```js
  const { data, error } = await storageClient.from('bucket').remove(['path/to/file'])
  ```

- Create signed URL to download file without requiring permissions:

  ```js
  const expireIn = 60
  const { data, error } = await storageClient
    .from('bucket')
    .createSignedUrl('path/to/file', expireIn)
  ```

- Retrieve URLs for assets in public buckets:

  ```js
  const { data, error } = await storageClient.from('public-bucket').getPublicUrl('path/to/file')
  ```

## Analytics Buckets

Supabase Storage provides specialized analytics buckets using Apache Iceberg table format, optimized for analytical workloads and large-scale data processing. These buckets are designed for data lake architectures, time-series data, and business intelligence applications.

### What are Analytics Buckets?

Analytics buckets use the Apache Iceberg open table format, providing:

- **ACID transactions** for data consistency
- **Schema evolution** without data rewrites
- **Time travel** to query historical data
- **Efficient metadata management** for large datasets
- **Optimized for analytical queries** rather than individual file operations

### When to Use Analytics Buckets

**Use analytics buckets for:**

- Time-series data (logs, metrics, events)
- Data lake architectures
- Business intelligence and reporting
- Large-scale batch processing
- Analytical workloads requiring ACID guarantees

**Use regular storage buckets for:**

- User file uploads (images, documents, videos)
- Individual file management
- Content delivery
- Simple object storage needs

### Quick Start

You can access analytics functionality through the `analytics` property on your storage client:

#### Via Supabase Client

```typescript
import { createClient } from '@supabase/supabase-js'
const supabase = createClient('https://your-project.supabase.co', 'your-publishable-key')
// Access analytics operations
const analytics = supabase.storage.analytics
// Create an analytics bucket
const { data, error } = await analytics.createBucket('analytics-data')
if (error) {
  console.error('Failed to create analytics bucket:', error.message)
} else {
  console.log('Created bucket:', data.name)
}
```

#### Via StorageClient

```typescript
import { StorageClient } from '@supabase/storage-js'
const storageClient = new StorageClient('https://your-project.supabase.co/storage/v1', {
  apikey: 'YOUR_API_KEY',
  Authorization: 'Bearer YOUR_TOKEN',
})
// Access analytics operations
const analytics = storageClient.analytics
// Create an analytics bucket
await analytics.createBucket('analytics-data')
```

### API Reference

#### Create Analytics Bucket

Creates a new analytics bucket using Iceberg table format:

```typescript
const { data, error } = await analytics.createBucket('my-analytics-bucket')
if (error) {
  console.error('Error:', error.message)
} else {
  console.log('Created bucket:', data)
}
```

**Returns:**

```typescript
{
  data: {
    id: string
    type: 'ANALYTICS'
    format: string
    created_at: string
    updated_at: string
  } | null
  error: StorageError | null
}
```

#### List Analytics Buckets

Retrieves all analytics buckets in your project with optional filtering and pagination:

```typescript
const { data, error } = await analytics.listBuckets({
  limit: 10,
  offset: 0,
  sortColumn: 'created_at',
  sortOrder: 'desc',
  search: 'prod',
})
if (data) {
  console.log(`Found ${data.length} analytics buckets`)
  data.forEach((bucket) => {
    console.log(`- ${bucket.id} (created: ${bucket.created_at})`)
  })
}
```

**Parameters:**

- `limit?: number` - Maximum number of buckets to return
- `offset?: number` - Number of buckets to skip (for pagination)
- `sortColumn?: 'id' | 'name' | 'created_at' | 'updated_at'` - Column to sort by
- `sortOrder?: 'asc' | 'desc'` - Sort direction
- `search?: string` - Search term to filter bucket names

**Returns:**

```typescript
{
  data: AnalyticBucket[] | null
  error: StorageError | null
}
```

**Example with Pagination:**

```typescript
// Fetch first page
const firstPage = await analytics.listBuckets({
  limit: 100,
  offset: 0,
  sortColumn: 'created_at',
  sortOrder: 'desc',
})
// Fetch second page
const secondPage = await analytics.listBuckets({
  limit: 100,
  offset: 100,
  sortColumn: 'created_at',
  sortOrder: 'desc',
})
```

#### Delete Analytics Bucket

Deletes an analytics bucket. The bucket must be empty before deletion.

```typescript
const { data, error } = await analytics.deleteBucket('old-analytics-bucket')
if (error) {
  console.error('Failed to delete:', error.message)
} else {
  console.log('Bucket deleted:', data.message)
}
```

**Returns:**

```typescript
{
  data: { message: string } | null
  error: StorageError | null
}
```

> **Note:** A bucket cannot be deleted if it contains data. You must empty the bucket first.

#### Get Iceberg Catalog for Advanced Operations

For advanced operations like creating tables, namespaces, and querying Iceberg metadata, use the `from()` method to get a configured [iceberg-js](https://github.com/supabase/iceberg-js) client:

```typescript
// Get an Iceberg REST Catalog client for your analytics bucket
const catalog = analytics.from('analytics-data')
// Create a namespace
await catalog.createNamespace({ namespace: ['default'] }, { properties: { owner: 'data-team' } })
// Create a table with schema
await catalog.createTable(
  { namespace: ['default'] },
  {
    name: 'events',
    schema: {
      type: 'struct',
      fields: [
        { id: 1, name: 'id', type: 'long', required: true },
        { id: 2, name: 'timestamp', type: 'timestamp', required: true },
        { id: 3, name: 'user_id', type: 'string', required: false },
      ],
      'schema-id': 0,
      'identifier-field-ids': [1],
    },
    'partition-spec': {
      'spec-id': 0,
      fields: [],
    },
    'write-order': {
      'order-id': 0,
      fields: [],
    },
    properties: {
      'write.format.default': 'parquet',
    },
  }
)
// List tables in namespace
const tables = await catalog.listTables({ namespace: ['default'] })
console.log(tables) // [{ namespace: ['default'], name: 'events' }]
// Load table metadata
const table = await catalog.loadTable({ namespace: ['default'], name: 'events' })
// Update table properties
await catalog.updateTable(
  { namespace: ['default'], name: 'events' },
  { properties: { 'read.split.target-size': '134217728' } }
)
// Drop table
await catalog.dropTable({ namespace: ['default'], name: 'events' })
// Drop namespace
await catalog.dropNamespace({ namespace: ['default'] })
```

**Returns:** `IcebergRestCatalog` instance from [iceberg-js](https://github.com/supabase/iceberg-js)

> **Note:** The `from()` method returns an Iceberg REST Catalog client that provides full access to the Apache Iceberg REST API. For complete documentation of available operations, see the [iceberg-js documentation](https://supabase.github.io/iceberg-js/).

### Error Handling

Analytics buckets use the same error handling pattern as the rest of the Storage SDK:

```typescript
const { data, error } = await analytics.createBucket('my-bucket')
if (error) {
  console.error('Error:', error.message)
  console.error('Status:', error.status)
  console.error('Status Code:', error.statusCode)
  // Handle error appropriately
}
```

#### Throwing Errors

You can configure the client to throw errors instead of returning them:

```typescript
const analytics = storageClient.analytics
analytics.throwOnError()
try {
  const { data } = await analytics.createBucket('my-bucket')
  // data is guaranteed to be present
  console.log('Success:', data)
} catch (error) {
  if (error instanceof StorageApiError) {
    console.error('API Error:', error.statusCode, error.message)
  }
}
```

### TypeScript Types

The library exports TypeScript types for analytics buckets:

```typescript
import type { AnalyticBucket, BucketType, StorageError } from '@supabase/storage-js'
// AnalyticBucket type
interface AnalyticBucket {
  id: string
  type: 'ANALYTICS'
  format: string
  created_at: string
  updated_at: string
}
```

### Common Patterns

#### Checking if a Bucket Exists

```typescript
async function bucketExists(bucketName: string): Promise<boolean> {
  const { data, error } = await analytics.listBuckets({
    search: bucketName,
  })
  if (error) {
    console.error('Error checking bucket:', error.message)
    return false
  }
  return data?.some((bucket) => bucket.id === bucketName) ?? false
}
```

#### Creating Bucket with Error Handling

```typescript
async function ensureAnalyticsBucket(bucketName: string) {
  // Try to create the bucket
  const { data, error } = await analytics.createBucket(bucketName)
  if (error) {
    // Check if bucket already exists (conflict error)
    if (error.statusCode === '409') {
      console.log(`Bucket '${bucketName}' already exists`)
      return { success: true, created: false }
    }
    // Other error occurred
    console.error('Failed to create bucket:', error.message)
    return { success: false, error }
  }
  console.log(`Created new bucket: '${bucketName}'`)
  return { success: true, created: true, data }
}
```

#### Listing All Buckets with Pagination

```typescript
async function getAllAnalyticsBuckets() {
  const allBuckets: AnalyticBucket[] = []
  let offset = 0
  const limit = 100
  while (true) {
    const { data, error } = await analytics.listBuckets({
      limit,
      offset,
      sortColumn: 'created_at',
      sortOrder: 'desc',
    })
    if (error) {
      console.error('Error fetching buckets:', error.message)
      break
    }
    if (!data || data.length === 0) {
      break
    }
    allBuckets.push(...data)
    // If we got fewer results than the limit, we've reached the end
    if (data.length < limit) {
      break
    }
    offset += limit
  }
  return allBuckets
}
```

## Vector Embeddings

Supabase Storage provides built-in support for storing and querying high-dimensional vector embeddings, powered by S3 Vectors. This enables semantic search, similarity matching, and AI-powered applications without needing a separate vector database.

> **Note:** Vector embeddings functionality is available in `@supabase/storage-js` v2.76 and later.

### Features

- **Vector Buckets**: Organize vector indexes into logical containers
- **Vector Indexes**: Define schemas with configurable dimensions and distance metrics
- **Batch Operations**: Insert/update/delete up to 500 vectors per request
- **Similarity Search**: Query for nearest neighbors using cosine, euclidean, or dot product distance
- **Metadata Filtering**: Store and filter vectors by arbitrary JSON metadata
- **Pagination**: Efficiently scan large vector datasets
- **Parallel Scanning**: Distribute scans across multiple workers for high throughput
- **Cross-platform**: Works in Node.js, browsers, and edge runtimes

### Quick Start

You can access vector functionality in three ways, depending on your use case:

#### Option 1: Via Supabase Client (Most Common)

If you're using the full Supabase client:

```typescript
import { createClient } from '@supabase/supabase-js'
const supabase = createClient('https://your-project.supabase.co', 'your-publishable-key')
// Access vector operations through storage
const vectors = supabase.storage.vectors
// Create a vector bucket
await vectors.createBucket('embeddings-prod')
// Create an index
const bucket = vectors.from('embeddings-prod')
await bucket.createIndex({
  indexName: 'documents-openai',
  dataType: 'float32',
  dimension: 1536,
  distanceMetric: 'cosine',
})
// Insert vectors
const index = bucket.index('documents-openai')
await index.putVectors({
  vectors: [
    {
      key: 'doc-1',
      data: { float32: [0.1, 0.2, 0.3 /* ...1536 dimensions */] },
      metadata: { title: 'Introduction', category: 'docs' },
    },
  ],
})
// Query similar vectors
const { data, error } = await index.queryVectors({
  queryVector: { float32: [0.15, 0.25, 0.35 /* ...1536 dimensions */] },
  topK: 5,
  returnDistance: true,
  returnMetadata: true,
})
if (data) {
  data.matches.forEach((match) => {
    console.log(`${match.key}: distance=${match.distance}`)
    console.log('Metadata:', match.metadata)
  })
}
```

#### Option 2: Via StorageClient

If you're using the standalone `StorageClient` for storage operations, access vectors through the `vectors` property:

```typescript
import { StorageClient } from '@supabase/storage-js'
const storageClient = new StorageClient('https://your-project.supabase.co/storage/v1', {
  apikey: 'YOUR_API_KEY',
  Authorization: 'Bearer YOUR_TOKEN',
})
// Access vector operations
const vectors = storageClient.vectors
// Use the same API as shown in Option 1
await vectors.createBucket('embeddings-prod')
const bucket = vectors.from('embeddings-prod')
// ... rest of operations
```

#### Option 3: Standalone Vector Client

For vector-only applications that don't need regular file storage operations:

```typescript
import { StorageVectorsClient } from '@supabase/storage-js'
// Initialize standalone vector client
const vectorClient = new StorageVectorsClient('https://your-project.supabase.co/storage/v1', {
  headers: { Authorization: 'Bearer YOUR_TOKEN' },
})
// Use the same API as shown in Option 1
await vectorClient.createBucket('embeddings-prod')
const bucket = vectorClient.from('embeddings-prod')
// ... rest of operations
```

> **When to use each approach:**
>
> - **Option 1**: When using other Supabase features (auth, database, realtime)
> - **Option 2**: When working with both file storage and vectors
> - **Option 3**: For dedicated vector-only applications without file storage

### API Reference

#### Client Initialization

```typescript
const vectorClient = new StorageVectorsClient(url, options?)
```

**Options:**

- `headers?: Record<string, string>` - Custom HTTP headers (e.g., Authorization)
- `fetch?: Fetch` - Custom fetch implementation

#### Vector Buckets

Vector buckets are top-level containers for organizing vector indexes.

##### Create Bucket

```typescript
const { data, error } = await vectorClient.createBucket('my-bucket')
```

##### Get Bucket

```typescript
const { data, error } = await vectorClient.getBucket('my-bucket')
console.log('Created at:', new Date(data.vectorBucket.creationTime! * 1000))
```

##### List Buckets

```typescript
const { data, error } = await vectorClient.listBuckets({
  prefix: 'prod-',
  maxResults: 100,
})
// Pagination
if (data?.nextToken) {
  const next = await vectorClient.listBuckets({ nextToken: data.nextToken })
}
```

##### Delete Bucket

```typescript
// Bucket must be empty (all indexes deleted first)
const { error } = await vectorClient.deleteBucket('my-bucket')
```

#### Vector Indexes

Vector indexes define the schema for embeddings including dimension and distance metric.

##### Create Index

```typescript
const bucket = vectorClient.from('my-bucket')
await bucket.createIndex({
  indexName: 'my-index',
  dataType: 'float32',
  dimension: 1536,
  distanceMetric: 'cosine', // 'cosine' | 'euclidean' | 'dotproduct'
  metadataConfiguration: {
    nonFilterableMetadataKeys: ['raw_text', 'internal_id'],
  },
})
```

**Distance Metrics:**

- `cosine` - Cosine similarity (normalized dot product)
- `euclidean` - Euclidean distance (L2 norm)
- `dotproduct` - Dot product similarity

##### Get Index

```typescript
const { data, error } = await bucket.getIndex('my-index')
console.log('Dimension:', data?.index.dimension)
console.log('Distance metric:', data?.index.distanceMetric)
```

##### List Indexes

```typescript
const { data, error } = await bucket.listIndexes({
  prefix: 'documents-',
  maxResults: 100,
})
```

##### Delete Index

```typescript
// Deletes index and all its vectors
await bucket.deleteIndex('my-index')
```

#### Vector Operations

##### Insert/Update Vectors (Upsert)

```typescript
const index = vectorClient.from('my-bucket').index('my-index')
await index.putVectors({
  vectors: [
    {
      key: 'unique-id-1',
      data: {
        float32: [
          /* 1536 numbers */
        ],
      },
      metadata: {
        title: 'Document Title',
        category: 'technical',
        page: 1,
      },
    },
    // ... up to 500 vectors per request
  ],
})
```

**Limitations:**

- 1-500 vectors per request
- Vectors must match index dimension
- Keys must be unique within index

##### Get Vectors by Key

```typescript
const { data, error } = await index.getVectors({
  keys: ['doc-1', 'doc-2', 'doc-3'],
  returnData: true, // Include embeddings
  returnMetadata: true, // Include metadata
})
data?.vectors.forEach((v) => {
  console.log(v.key, v.metadata)
})
```

##### Query Similar Vectors (ANN Search)

```typescript
const { data, error } = await index.queryVectors({
  queryVector: {
    float32: [
      /* 1536 numbers */
    ],
  },
  topK: 10,
  filter: {
    category: 'technical',
    published: true,
  },
  returnDistance: true,
  returnMetadata: true,
})
// Results ordered by similarity
data?.matches.forEach((match) => {
  console.log(`${match.key}: distance=${match.distance}`)
})
```

**Filter Syntax:**
The `filter` parameter accepts arbitrary JSON for metadata filtering. Non-filterable keys (configured at index creation) cannot be used in filters but can still be returned.

##### List/Scan Vectors

```typescript
// Simple pagination
let nextToken: string | undefined
do {
  const { data } = await index.listVectors({
    maxResults: 500,
    nextToken,
    returnMetadata: true,
  })
  console.log('Batch:', data?.vectors.length)
  nextToken = data?.nextToken
} while (nextToken)
// Parallel scanning (4 workers)
const workers = [0, 1, 2, 3].map(async (segmentIndex) => {
  const { data } = await index.listVectors({
    segmentCount: 4,
    segmentIndex,
    returnMetadata: true,
  })
  return data?.vectors || []
})
const results = await Promise.all(workers)
const allVectors = results.flat()
```

**Limitations:**

- `maxResults`: 1-1000 (default: 500)
- `segmentCount`: 1-16
- Response may be limited by 1MB size

##### Delete Vectors

```typescript
await index.deleteVectors({
  keys: ['doc-1', 'doc-2', 'doc-3'],
  // ... up to 500 keys per request
})
```

### Error Handling

The library uses a consistent error handling pattern:

```typescript
const { data, error } = await vectorClient.createBucket('my-bucket')
if (error) {
  console.error('Error:', error.message)
  console.error('Status:', error.status)
  console.error('Code:', error.statusCode)
}
```

#### Error Codes

| Code                         | HTTP | Description             |
| ---------------------------- | ---- | ----------------------- |
| `InternalError`              | 500  | Internal server error   |
| `S3VectorConflictException`  | 409  | Resource already exists |
| `S3VectorNotFoundException`  | 404  | Resource not found      |
| `S3VectorBucketNotEmpty`     | 400  | Bucket contains indexes |
| `S3VectorMaxBucketsExceeded` | 400  | Bucket quota exceeded   |
| `S3VectorMaxIndexesExceeded` | 400  | Index quota exceeded    |

#### Throwing Errors

You can configure the client to throw errors instead:

```typescript
const vectorClient = new StorageVectorsClient(url, options)
vectorClient.throwOnError()
try {
  const { data } = await vectorClient.createBucket('my-bucket')
  // data is guaranteed to be present
} catch (error) {
  if (error instanceof StorageVectorsApiError) {
    console.error('API Error:', error.statusCode)
  }
}
```

### Advanced Usage

#### Scoped Clients

Create scoped clients for cleaner code:

```typescript
// Bucket-scoped operations
const bucket = vectorClient.from('embeddings-prod')
await bucket.createIndex({
  /* ... */
})
await bucket.listIndexes()
// Index-scoped operations
const index = bucket.index('documents-openai')
await index.putVectors({
  /* ... */
})
await index.queryVectors({
  /* ... */
})
```

#### Custom Fetch

Provide a custom fetch implementation:

```typescript
import { StorageVectorsClient } from '@supabase/storage-js'
const vectorClient = new StorageVectorsClient(url, {
  fetch: customFetch,
  headers: {
    /* ... */
  },
})
```

#### Batch Processing

Process large datasets in batches:

```typescript
async function insertLargeDataset(vectors: VectorObject[]) {
  const batchSize = 500
  for (let i = 0; i < vectors.length; i += batchSize) {
    const batch = vectors.slice(i, i + batchSize)
    await index.putVectors({ vectors: batch })
    console.log(`Inserted ${i + batch.length}/${vectors.length}`)
  }
}
```

#### Float32 Validation

Ensure vectors are properly normalized to float32:

```typescript
import { normalizeToFloat32 } from '@supabase/storage-js'
const vector = normalizeToFloat32([0.1, 0.2, 0.3 /* ... */])
```

### Type Definitions

The library exports comprehensive TypeScript types:

```typescript
import type {
  VectorBucket,
  VectorIndex,
  VectorData,
  VectorObject,
  VectorMatch,
  VectorMetadata,
  DistanceMetric,
  ApiResponse,
  StorageVectorsError,
} from '@supabase/storage-js'
```

## Development

This package is part of the [Supabase JavaScript monorepo](https://github.com/supabase/supabase-js). To work on this package:

### Building

#### Build Scripts Overview

```bash
# Build the package
pnpm nx build storage-js
# Watch mode for development
pnpm nx build storage-js --watch
# Generate documentation
pnpm nx docs storage-js
```

### Testing

**Important:** The storage-js tests require a local Supabase stack running via the Supabase CLI. Docker must be running since the Supabase CLI uses it internally.

#### Prerequisites

1. **Docker** must be installed and running (used by Supabase CLI internally)
2. **Supabase CLI** — installed automatically via `pnpm exec supabase`

#### Test Scripts Overview

| Script            | Description                       | What it does                                                      |
| ----------------- | --------------------------------- | ----------------------------------------------------------------- |
| `test:storage`    | **Complete test workflow**        | Runs the full test cycle: clean → start infra → run tests → clean |
| `test:suite`      | **Jest tests only**               | Runs Jest tests with coverage (requires infra to be running)      |
| `test:infra`      | **Start test infrastructure**     | Starts Supabase CLI stack (PostgreSQL, Storage API, Kong, etc.)   |
| `test:clean-post` | **Stop and clean infrastructure** | Stops the Supabase CLI stack                                      |

#### Running Tests

##### Option 1: Complete Test Run (Recommended)

This handles everything automatically - starting infrastructure, running tests, and cleaning up:

```bash
# From monorepo root
pnpm nx test:storage storage-js
```

This command will:

1. Stop any existing test containers
2. Build and start fresh test infrastructure
3. Wait for services to be ready
4. Run all Jest tests with coverage
5. Clean up all containers after tests complete

##### Option 2: Manual Infrastructure Management

Useful for development when you want to run tests multiple times without restarting Docker:

```bash
# Step 1: Start the test infrastructure
# From root
pnpm nx test:infra storage-js
# This starts: PostgreSQL, Storage API, Kong Gateway, and imgproxy
# Step 2: Run tests (can run multiple times)
pnpm nx test:suite storage-js
# Step 3: When done, clean up the infrastructure
pnpm nx test:clean-post storage-js
```

##### Option 3: Development Mode

For actively developing and debugging tests:

```bash
# Start infrastructure once (from root)
pnpm nx test:infra storage-js
# Run tests in watch mode
pnpm nx test:suite storage-js --watch
# Clean up when done
pnpm nx test:clean-post storage-js
```

#### Test Infrastructure Details

The test infrastructure is managed via the Supabase CLI (`pnpm exec supabase start --workdir test`), which starts a local Supabase stack defined by the config in `test/`. This includes PostgreSQL, the Storage API, Kong Gateway, and supporting services.

#### Common Issues and Solutions

| Issue                             | Solution                                                                                          |
| --------------------------------- | ------------------------------------------------------------------------------------------------- |
| Port conflicts                    | Another service is using a required port. Run `pnpm nx test:clean-post storage-js` then try again |
| "request failed, reason:" errors  | Infrastructure isn't running. Run `pnpm nx test:infra storage-js` first                           |
| Tests fail with connection errors | Ensure Docker is running (Supabase CLI requires Docker)                                           |
| Stack already running             | Run `pnpm nx test:clean-post storage-js` to stop it before restarting                             |

#### Understanding Test Failures

- **StorageUnknownError with "request failed"**: Infrastructure not running
- **Snapshot failures**: Expected test data has changed — review and update snapshots if needed

### Contributing

We welcome contributions! Please see our [Contributing Guide](../../../CONTRIBUTING.md) for details on how to get started.

For major changes or if you're unsure about something, please open an issue first to discuss your proposed changes.