1# Features & Capabilities
2 
3## Caching
4 
5Dashboard: Settings → Cache Responses → Enable
6 
7```typescript
8// Custom TTL (1 hour)
9headers: { 'cf-aig-cache-ttl': '3600' }
10 
11// Skip cache
12headers: { 'cf-aig-skip-cache': 'true' }
13 
14// Custom cache key
15headers: { 'cf-aig-cache-key': 'greeting-en' }
16```
17 
18**Limits:** TTL 60s - 30 days. **Does NOT work with streaming.**
19 
20## Rate Limiting
21 
22Dashboard: Settings → Rate-limiting → Enable
23 
24- **Fixed window:** Resets at intervals
25- **Sliding window:** Rolling window (more accurate)
26- Returns `429` when exceeded
27 
28## Guardrails
29 
30Dashboard: Settings → Guardrails → Enable
31 
32Filter prompts/responses for inappropriate content. Actions: Flag (log) or Block (reject).
33 
34## Data Loss Prevention (DLP)
35 
36Dashboard: Settings → DLP → Enable
37 
38Detect PII (emails, SSNs, credit cards). Actions: Flag, Block, or Redact.
39 
40## Billing Modes
41 
42| Mode | Description | Setup |
43|------|-------------|-------|
44| **Unified Billing** | Pay through Cloudflare, no provider keys | Use `cf-aig-authorization` header only |
45| **BYOK** | Store provider keys in dashboard | Add keys in Provider Keys section |
46| **Pass-through** | Send provider key with each request | Include provider's auth header |
47 
48## Zero Data Retention
49 
50Dashboard: Settings → Privacy → Zero Data Retention
51 
52No prompts/responses stored. Request counts and costs still tracked.
53 
54## Logging
55 
56Dashboard: Settings → Logs → Enable (up to 10M logs)
57 
58Each entry: prompt, response, provider, model, tokens, cost, duration, cache status, metadata.
59 
60```typescript
61// Skip logging for request
62headers: { 'cf-aig-collect-log': 'false' }
63```
64 
65**Export:** Use Logpush to S3, GCS, Datadog, Splunk, etc.
66 
67## Custom Cost Tracking
68 
69For models not in Cloudflare's pricing database:
70 
71Dashboard: Gateway → Settings → Custom Costs
72 
73Or via API: set `model`, `input_cost`, `output_cost`.
74 
75## Supported Providers (22+)
76 
77| Provider | Unified API | Notes |
78|----------|-------------|-------|
79| OpenAI | `openai/gpt-4o` | Full support |
80| Anthropic | `anthropic/claude-sonnet-4-5` | Full support |
81| Google AI | `google-ai-studio/gemini-2.0-flash` | Full support |
82| Workers AI | `workersai/@cf/meta/llama-3` | Native |
83| Azure OpenAI | `azure-openai/*` | Deployment names |
84| AWS Bedrock | Provider endpoint only | `/bedrock/*` |
85| Groq | `groq/*` | Fast inference |
86| Mistral, Cohere, Perplexity, xAI, DeepSeek, Cerebras | Full support | - |
87 
88## Best Practices
89 
901. Enable caching for deterministic prompts
912. Set rate limits to prevent abuse
923. Use guardrails for user-facing AI
934. Enable DLP for sensitive data
945. Use unified billing or BYOK for simpler key management
956. Enable logging for debugging
967. Use zero data retention when privacy required
97