1# Vectorize Patterns
2 
3## Workers AI Integration
4 
5```typescript
6// Generate embedding + query
7const result = await env.AI.run("@cf/baai/bge-base-en-v1.5", { text: [query] });
8const matches = await env.VECTORIZE.query(result.data[0], { topK: 5 }); // Pass data[0]!
9```
10 
11| Model | Dimensions |
12|-------|------------|
13| `@cf/baai/bge-small-en-v1.5` | 384 |
14| `@cf/baai/bge-base-en-v1.5` | 768 (recommended) |
15| `@cf/baai/bge-large-en-v1.5` | 1024 |
16 
17## OpenAI Integration
18 
19```typescript
20const response = await openai.embeddings.create({ model: "text-embedding-ada-002", input: query });
21const matches = await env.VECTORIZE.query(response.data[0].embedding, { topK: 5 });
22```
23 
24## RAG Pattern
25 
26```typescript
27// 1. Embed query
28const emb = await env.AI.run("@cf/baai/bge-base-en-v1.5", { text: [query] });
29 
30// 2. Search vectors
31const matches = await env.VECTORIZE.query(emb.data[0], { topK: 5, returnMetadata: "indexed" });
32 
33// 3. Fetch full docs from R2/D1/KV
34const docs = await Promise.all(matches.matches.map(m => env.R2.get(m.metadata.key).then(o => o?.text())));
35 
36// 4. Generate with context
37const answer = await env.AI.run("@cf/meta/llama-3-8b-instruct", {
38  prompt: `Context:\n${docs.filter(Boolean).join("\n\n")}\n\nQuestion: ${query}\n\nAnswer:`
39});
40```
41 
42## Multi-Tenant
43 
44### Namespaces (< 50K tenants, fastest)
45 
46```typescript
47await env.VECTORIZE.upsert([{ id: "1", values: emb, namespace: `tenant-${id}` }]);
48await env.VECTORIZE.query(vec, { namespace: `tenant-${id}`, topK: 10 });
49```
50 
51### Metadata Filter (> 50K tenants)
52 
53```bash
54wrangler vectorize create-metadata-index my-index --property-name=tenantId --type=string
55```
56 
57```typescript
58await env.VECTORIZE.upsert([{ id: "1", values: emb, metadata: { tenantId: id } }]);
59await env.VECTORIZE.query(vec, { filter: { tenantId: id }, topK: 10 });
60```
61 
62## Hybrid Search
63 
64```typescript
65const matches = await env.VECTORIZE.query(vec, {
66  topK: 20,
67  filter: {
68    category: { $in: ["tech", "science"] },
69    published: { $gte: lastMonthTimestamp }
70  }
71});
72```
73 
74## Batch Ingestion
75 
76```typescript
77const BATCH = 500;
78for (let i = 0; i < vectors.length; i += BATCH) {
79  await env.VECTORIZE.upsert(vectors.slice(i, i + BATCH));
80}
81```
82 
83## Best Practices
84 
851. **Pass `data[0]`** not `data` or full response
862. **Batch 500** vectors per upsert
873. **Create metadata indexes** before inserting
884. **Use namespaces** for tenant isolation (faster than filters)
895. **`returnMetadata: "indexed"`** for best speed/data balance
906. **Handle 5-10s mutation delay** in async operations
91