1# AI Search Patterns
2 
3## search() vs aiSearch()
4 
5| Use | Method | Returns |
6|-----|--------|---------|
7| Custom UI, analytics | `search()` | Raw chunks only (~100-300ms) |
8| Chatbots, Q&A | `aiSearch()` | AI response + chunks (~500-2000ms) |
9 
10## rewrite_query
11 
12| Setting | Use When |
13|---------|----------|
14| `true` | User input (typos, vague queries) |
15| `false` | LLM-generated queries (already optimized) |
16 
17## Multitenancy (Folder-Based)
18 
19```typescript
20const answer = await env.AI.autorag("saas-docs").aiSearch({
21  query: "refund policy",
22  model: "@cf/meta/llama-3.3-70b-instruct-fp8-fast",
23  filters: {
24    column: "folder",
25    operator: "gte",  // "starts with" pattern
26    value: `tenants/${tenantId}/`
27  }
28});
29```
30 
31## Streaming
32 
33```typescript
34const stream = await env.AI.autorag("docs").aiSearch({
35  query, model: "@cf/meta/llama-3.3-70b-instruct-fp8-fast", stream: true
36});
37return new Response(stream, { headers: { "Content-Type": "text/event-stream" } });
38```
39 
40## Score Threshold
41 
42| Threshold | Use |
43|-----------|-----|
44| 0.3 (default) | Broad recall, exploratory |
45| 0.5 | Balanced, production default |
46| 0.7 | High precision, critical accuracy |
47 
48## System Prompt Template
49 
50```typescript
51const systemPrompt = `You are a documentation assistant.
52- Answer ONLY based on provided context
53- If context doesn't contain answer, say "I don't have information"
54- Include code examples from context`;
55```
56 
57## Compound Filters
58 
59```typescript
60// OR: Multiple folders
61filters: {
62  operator: "or",
63  filters: [
64    { column: "folder", operator: "gte", value: "docs/api/" },
65    { column: "folder", operator: "gte", value: "docs/auth/" }
66  ]
67}
68 
69// AND: Folder + date
70filters: {
71  operator: "and",
72  filters: [
73    { column: "folder", operator: "gte", value: "docs/" },
74    { column: "timestamp", operator: "gte", value: oneWeekAgoSeconds }
75  ]
76}
77```
78 
79## Reranking
80 
81Enable for high-stakes use cases (adds ~300ms latency):
82 
83```typescript
84reranking: { enabled: true, model: "@cf/baai/bge-reranker-base" }
85```
86