1# Cloudflare AI Search Reference
2 
3Expert guidance for implementing Cloudflare AI Search (formerly AutoRAG), Cloudflare's managed semantic search and RAG service.
4 
5## Overview
6 
7**AI Search** is a managed RAG (Retrieval-Augmented Generation) pipeline that combines:
8- Automatic semantic indexing of your content
9- Vector similarity search
10- Built-in LLM generation
11 
12**Key value propositions:**
13- **Zero vector management** - No manual embedding, indexing, or storage
14- **Auto-indexing** - Content automatically re-indexed every 6 hours
15- **Built-in generation** - Optional AI response generation from retrieved context
16- **Multi-source** - Index from R2 buckets or website crawls
17 
18**Data source options:**
19- **R2 bucket** - Index files from Cloudflare R2 (supports MD, TXT, HTML, PDF, DOC, CSV, JSON)
20- **Website** - Crawl and index website content (requires Cloudflare-hosted domain)
21 
22**Indexing lifecycle:**
23- Automatic 6-hour refresh cycle
24- Manual "Force Sync" available (30s rate limit)
25- Not designed for real-time updates
26 
27## Quick Start
28 
29**1. Create AI Search instance in dashboard:**
30- Go to Cloudflare Dashboard → AI Search → Create
31- Choose data source (R2 or website)
32- Configure instance name and settings
33 
34**2. Configure Worker:**
35 
36```jsonc
37// wrangler.jsonc
38{
39  "ai": {
40    "binding": "AI"
41  }
42}
43```
44 
45**3. Use in Worker:**
46 
47```typescript
48export default {
49  async fetch(request, env) {
50    const answer = await env.AI.autorag("my-search-instance").aiSearch({
51      query: "How do I configure caching?",
52      model: "@cf/meta/llama-3.3-70b-instruct-fp8-fast"
53    });
54    
55    return Response.json({ answer: answer.response });
56  }
57};
58```
59 
60## When to Use AI Search
61 
62### AI Search vs Vectorize
63 
64| Factor | AI Search | Vectorize |
65|--------|-----------|-----------|
66| **Management** | Fully managed | Manual embedding + indexing |
67| **Use when** | Want zero-ops RAG pipeline | Need custom embeddings/control |
68| **Indexing** | Automatic (6hr cycle) | Manual via API |
69| **Generation** | Built-in optional | Bring your own LLM |
70| **Data sources** | R2 or website | Manual insert |
71| **Best for** | Docs, support, enterprise search | Custom ML pipelines, real-time |
72 
73### AI Search vs Direct Workers AI
74 
75| Factor | AI Search | Workers AI (direct) |
76|--------|-----------|---------------------|
77| **Context** | Automatic retrieval | Manual context building |
78| **Use when** | Need RAG (search + generate) | Simple generation tasks |
79| **Indexing** | Built-in | Not applicable |
80| **Best for** | Knowledge bases, docs | Simple chat, transformations |
81 
82### search() vs aiSearch()
83 
84| Method | Returns | Use When |
85|--------|---------|----------|
86| `search()` | Search results only | Building custom UI, need raw chunks |
87| `aiSearch()` | AI response + results | Need ready-to-use answer (chatbot, Q&A) |
88 
89### Real-time Updates Consideration
90 
91**AI Search is NOT ideal if:**
92- Need real-time content updates (<6 hours)
93- Content changes multiple times per hour
94- Strict freshness requirements
95 
96**AI Search IS ideal if:**
97- Content relatively stable (docs, policies, knowledge bases)
98- 6-hour refresh acceptable
99- Prefer zero-ops over real-time
100 
101## Platform Limits
102 
103| Limit | Value |
104|-------|-------|
105| Max instances per account | 10 |
106| Max files per instance | 100,000 |
107| Max file size | 4 MB |
108| Index frequency | Every 6 hours |
109| Force Sync rate limit | Once per 30 seconds |
110| Filter nesting depth | 2 levels |
111| Filters per compound | 10 |
112| Score threshold range | 0.0 - 1.0 |
113 
114## Reading Order
115 
116Navigate these references based on your task:
117 
118| Task | Read | Est. Time |
119|------|------|-----------|
120| **Understand AI Search** | README only | 5 min |
121| **Implement basic search** | README → api.md | 10 min |
122| **Configure data source** | README → configuration.md | 10 min |
123| **Production patterns** | patterns.md | 15 min |
124| **Debug issues** | gotchas.md | 10 min |
125| **Full implementation** | README → api.md → patterns.md | 30 min |
126 
127## In This Reference
128 
129- **[api.md](api.md)** - API endpoints, methods, TypeScript interfaces
130- **[configuration.md](configuration.md)** - Setup, data sources, wrangler config
131- **[patterns.md](patterns.md)** - Common patterns, decision guidance, code examples
132- **[gotchas.md](gotchas.md)** - Troubleshooting, code-level gotchas, limits
133 
134## See Also
135 
136- [Cloudflare AI Search Docs](https://developers.cloudflare.com/ai-search/)
137- [Workers AI Docs](https://developers.cloudflare.com/workers-ai/)
138- [Vectorize Docs](https://developers.cloudflare.com/vectorize/)
139