Skip to main content

Storing Vectors

还没翻译译自英文（supabase.com）本页 38 段都还没译

This feature is in alpha

Expect rapid changes, limited features, and possible breaking updates. Share feedback as we refine the experience and expand access.

Once you've created a bucket and index, you can start storing vectors. Vectors can include optional metadata for filtering and enrichment during queries.

Basic vector insertion

import { createClient } from '@supabase/supabase-js'

const supabase = createClient('https://your-project-id.supabase.co', 'your-service-key')

// Get bucket and index
const bucket = supabase.storage.vectors.from('embeddings')
const index = bucket.index('documents-openai')

// Insert vectors
const { error } = await index.putVectors({
  vectors: [
    {
      key: 'doc-1',
      data: {
        float32: [0.1, 0.2, 0.3 /* ... rest of embedding ... */],
      },
      metadata: {
        title: 'Getting Started with Vector Buckets',
        source: 'documentation',
      },
    },
    {
      key: 'doc-2',
      data: {
        float32: [0.4, 0.5, 0.6 /* ... rest of embedding ... */],
      },
      metadata: {
        title: 'Advanced Vector Search',
        source: 'blog',
      },
    },
  ],
})

if (error) {
  console.error('Error storing vectors:', error)
} else {
  console.log('✓ Vectors stored successfully')
}
from supabase import create_client

supabase = create_client('https://your-project-id.supabase.co', 'your-service-key')

# Get bucket and index
bucket = supabase.storage.vectors().from_('embeddings')
index = bucket.index('documents-openai')

# Insert vectors
index.put([
    {
        'key': 'doc-1',
        'data': {'float32': [0.1, 0.2, 0.3]},  # ... rest of embedding
        'metadata': {
            'title': 'Getting Started with Vector Buckets',
            'source': 'documentation',
        },
    },
    {
        'key': 'doc-2',
        'data': {'float32': [0.4, 0.5, 0.6]},  # ... rest of embedding
        'metadata': {
            'title': 'Advanced Vector Search',
            'source': 'blog',
        },
    },
])

print('✓ Vectors stored successfully')
-- Setup S3 Vector Wrapper (one-time setup)
-- https://supabase.com/docs/guides/database/extensions/wrappers/s3_vectors

-- Insert vectors into the index
INSERT INTO s3_vectors.documents_openai (key, data, metadata)
VALUES
  (
    'doc-1',
    '[0.1, 0.2, 0.3, /* ... rest of embedding ... */]'::embd,
    '{"title": "Getting Started with Vector Buckets", "source": "documentation"}'::jsonb
  ),
  (
    'doc-2',
    '[0.4, 0.5, 0.6, /* ... rest of embedding ... */]'::embd,
    '{"title": "Advanced Vector Search", "source": "blog"}'::jsonb
  );

Storing vectors from Embeddings API

Generate embeddings using an LLM API and store them directly:

import { createClient } from '@supabase/supabase-js'
import OpenAI from 'openai'

const supabase = createClient('https://your-project-id.supabase.co', 'your-service-key')

const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
})

// Documents to embed and store
const documents = [
  { id: '1', title: 'How to Train Your AI', content: 'Guide for training models...' },
  { id: '2', title: 'Vector Search Best Practices', content: 'Tips for semantic search...' },
  {
    id: '3',
    title: 'Building RAG Systems',
    content: 'Implementing retrieval-augmented generation...',
  },
]

// Generate embeddings
const embeddings = await openai.embeddings.create({
  model: 'text-embedding-3-small',
  input: documents.map((doc) => doc.content),
})

// Prepare vectors for storage
const vectors = documents.map((doc, index) => ({
  key: doc.id,
  data: {
    float32: embeddings.data[index].embedding,
  },
  metadata: {
    title: doc.title,
    source: 'knowledge_base',
    created_at: new Date().toISOString(),
  },
}))

// Store vectors in batches (max 500 per request)
const bucket = supabase.storage.vectors.from('embeddings')
const vectorIndex = bucket.index('documents-openai')

for (let i = 0; i < vectors.length; i += 500) {
  const batch = vectors.slice(i, i + 500)
  const { error } = await vectorIndex.putVectors({ vectors: batch })

  if (error) {
    console.error(`Error storing batch ${i / 500 + 1}:`, error)
  } else {
    console.log(`✓ Stored batch ${i / 500 + 1} (${batch.length} vectors)`)
  }
}
from supabase import create_client
from openai import OpenAI
from datetime import datetime

supabase = create_client('https://your-project-id.supabase.co', 'your-service-key')
openai = OpenAI(api_key=os.environ.get('OPENAI_API_KEY'))

# Documents to embed and store
documents = [
    {'id': '1', 'title': 'How to Train Your AI', 'content': 'Guide for training models...'},
    {'id': '2', 'title': 'Vector Search Best Practices', 'content': 'Tips for semantic search...'},
    {'id': '3', 'title': 'Building RAG Systems', 'content': 'Implementing retrieval-augmented generation...'},
]

# Generate embeddings
embeddings_response = openai.embeddings.create(
    model='text-embedding-3-small',
    input=[doc['content'] for doc in documents]
)

# Prepare vectors for storage
vectors = []
for doc, embedding_data in zip(documents, embeddings_response.data):
    vectors.append({
        'key': doc['id'],
        'data': {'float32': embedding_data.embedding},
        'metadata': {
            'title': doc['title'],
            'source': 'knowledge_base',
            'created_at': datetime.now().isoformat(),
        },
    })

# Store vectors in batches (max 500 per request)
bucket = supabase.storage.vectors().from_('embeddings')
vector_index = bucket.index('documents-openai')

batch_size = 500
for i in range(0, len(vectors), batch_size):
    batch = vectors[i:i + batch_size]
    vector_index.put(batch)
    print(f'✓ Stored batch {i // batch_size + 1} ({len(batch)} vectors)')
-- Insert vectors with pre-generated embeddings
INSERT INTO s3_vectors.documents_openai (key, data, metadata)
VALUES
  (
    '1',
    '[0.1, 0.2, 0.3, /* ... rest of embedding ... */]'::embd,
    '{"title": "How to Train Your AI", "source": "knowledge_base", "created_at": "2025-01-01T00:00:00Z"}'::jsonb
  ),
  (
    '2',
    '[0.4, 0.5, 0.6, /* ... rest of embedding ... */]'::embd,
    '{"title": "Vector Search Best Practices", "source": "knowledge_base", "created_at": "2025-01-01T00:00:00Z"}'::jsonb
  ),
  (
    '3',
    '[0.7, 0.8, 0.9, /* ... rest of embedding ... */]'::embd,
    '{"title": "Building RAG Systems", "source": "knowledge_base", "created_at": "2025-01-01T00:00:00Z"}'::jsonb
  );

Updating vectors

const index = bucket.index('documents-openai')

// Update a vector (same key)
const { error } = await index.putVectors({
  vectors: [
    {
      key: 'doc-1',
      data: {
        float32: [0.15, 0.25, 0.35 /* ... updated embedding ... */],
      },
      metadata: {
        title: 'Getting Started with Vector Buckets - Updated',
        updated_at: new Date().toISOString(),
      },
    },
  ],
})

if (!error) {
  console.log('✓ Vector updated successfully')
}
index = bucket.index('documents-openai')

# Update a vector (same key)
index.put([
    {
        'key': 'doc-1',
        'data': {'float32': [0.15, 0.25, 0.35]},  # ... updated embedding
        'metadata': {
            'title': 'Getting Started with Vector Buckets - Updated',
            'updated_at': datetime.now().isoformat(),
        },
    },
])

print('✓ Vector updated successfully')
-- Update a vector (delete and re-insert)
DELETE FROM s3_vectors.documents_openai WHERE key = 'doc-1';

INSERT INTO s3_vectors.documents_openai (key, data, metadata)
VALUES (
  'doc-1',
  '[0.15, 0.25, 0.35, /* ... updated embedding ... */]'::embd,
  '{"title": "Getting Started with Vector Buckets - Updated", "updated_at": "2025-01-01T00:00:00Z"}'::jsonb
);

Deleting vectors

const index = bucket.index('documents-openai')

// Delete specific vectors
const { error } = await index.deleteVectors({
  keys: ['doc-1', 'doc-2'],
})

if (!error) {
  console.log('✓ Vectors deleted successfully')
}
index = bucket.index('documents-openai')

# Delete specific vectors
index.delete(['doc-1', 'doc-2'])

print('✓ Vectors deleted successfully')
-- Delete vectors by key
delete from s3_vectors.documents_openai
where key in ('doc-1', 'doc-2');

Metadata best practices

Metadata makes vectors more useful by enabling filtering and context:

const vectors = [
  {
    key: 'product-001',
    data: { float32: [...] },
    metadata: {
      product_id: 'prod-001',
      category: 'electronics',
      price: 299.99,
      in_stock: true,
      tags: ['laptop', 'portable'],
      description: 'High-performance ultrabook'
    }
  },
  {
    key: 'product-002',
    data: { float32: [...] },
    metadata: {
      product_id: 'prod-002',
      category: 'electronics',
      price: 99.99,
      in_stock: true,
      tags: ['headphones', 'wireless'],
      description: 'Noise-cancelling wireless headphones'
    }
  }
]

const { error } = await index.putVectors({ vectors })

Metadata field guidelines

  • Keep it lightweight - Metadata is returned with query results, so large values increase response size
  • Use consistent types - Store the same field with consistent data types across vectors
  • Index key fields - Mark fields you'll filter by to improve query performance
  • Avoid nested objects - While supported, flat structures are easier to filter

Batch processing large datasets

For storing large numbers of vectors efficiently:

import { createClient } from '@supabase/supabase-js'
import fs from 'fs'

const supabase = createClient(...)
const index = supabase.storage.vectors
  .from('embeddings')
  .index('documents-openai')

// Read embeddings from file
const embeddingsFile = fs.readFileSync('embeddings.jsonl', 'utf-8')
const lines = embeddingsFile.split('\n').filter(line => line.trim())

const vectors = lines.map((line, idx) => {
  const { key, embedding, metadata } = JSON.parse(line)
  return {
    key,
    data: { float32: embedding },
    metadata
  }
})

// Process in batches
const BATCH_SIZE = 500
let processed = 0

for (let i = 0; i < vectors.length; i += BATCH_SIZE) {
  const batch = vectors.slice(i, i + BATCH_SIZE)

  try {
    const { error } = await index.putVectors({ vectors: batch })

    if (error) throw error

    processed += batch.length
    console.log(`Progress: ${processed}/${vectors.length}`)
  } catch (error) {
    console.error(`Batch failed at offset ${i}:`, error)
    // Optionally implement retry logic
  }
}

console.log('✓ All vectors stored successfully')
from supabase import create_client
import json

supabase = create_client(...)
index = supabase.storage.vectors().from_('embeddings').index('documents-openai')

# Read embeddings from file
with open('embeddings.jsonl', 'r') as f:
    lines = [line.strip() for line in f if line.strip()]

vectors = []
for line in lines:
    data = json.loads(line)
    vectors.append({
        'key': data['key'],
        'data': {'float32': data['embedding']},
        'metadata': data.get('metadata', {})
    })

# Process in batches
BATCH_SIZE = 500
processed = 0

for i in range(0, len(vectors), BATCH_SIZE):
    batch = vectors[i:i + BATCH_SIZE]

    try:
        index.put(batch)
        processed += len(batch)
        print(f'Progress: {processed}/{len(vectors)}')
    except Exception as e:
        print(f'Batch failed at offset {i}: {e}')
        # Optionally implement retry logic

print('✓ All vectors stored successfully')

Performance optimization

Batch operations

Always use batch operations for better performance:

// ❌ Inefficient - Multiple requests
for (const vector of vectors) {
  await index.putVectors({ vectors: [vector] })
}

// ✅ Efficient - Single batch operation
await index.putVectors({ vectors })
# ❌ Inefficient - Multiple requests
for vector in vectors:
    index.put([vector])

# ✅ Efficient - Single batch operation
index.put(vectors)

Metadata considerations

Keep metadata concise:

// ❌ Large metadata
metadata: {
  full_document_text: 'Very long document content...',
  detailed_analysis: { /* large object */ }
}

// ✅ Lean metadata
metadata: {
  doc_id: 'doc-123',
  category: 'news',
  summary: 'Brief summary'
}

Next steps

Edit on GitHub
AI 翻译查看英文原文