Cursor rule

1---
2description: "Core AI development patterns and best practices for the Megarepo AI setup repository"
3globs: ["src/**/*", "app/**/*", "pages/**/*"]
4alwaysApply: true
5---
6 
7# AI Development Best Practices
8 
9You are working on an AI-focused development project. Follow these core principles for AI integration:
10 
11## AI-First Development Principles
12 
13### 1. API Integration Patterns
14- Use modern async/await patterns for all AI API calls
15- Implement comprehensive error handling for AI service failures
16- Consider token usage and model performance optimization
17- Design for scalability with AI workloads
18 
19### 2. Preferred AI Integration Pattern
20```javascript
21const aiResponse = await fetch('/api/ai-service', {
22  method: 'POST',
23  headers: {
24    'Content-Type': 'application/json',
25    'Authorization': `Bearer ${process.env.AI_API_KEY}`
26  },
27  body: JSON.stringify({ prompt, options })
28});
29 
30if (!aiResponse.ok) {
31  throw new Error(`AI service error: ${aiResponse.status}`);
32}
33```
34 
35### 3. Recommended AI Dependencies
36When adding AI functionality, prefer these established packages:
37- `openai` - Official OpenAI API client
38- `@langchain/core` - LangChain framework
39- `@vercel/ai` - Vercel AI SDK
40- `@huggingface/inference` - Hugging Face API client
41 
42### 4. Response Handling Best Practices
43- Always validate AI responses before using them
44- Implement fallback mechanisms for failed requests
45- Log AI interactions for debugging and monitoring
46- Handle streaming responses appropriately
47 
48### 5. Performance Optimization
49- Cache AI responses when appropriate
50- Implement request debouncing for user inputs
51- Use background processing for long-running AI tasks
52- Consider edge functions for AI API proxying
53 
54### 6. User Experience Guidelines
55- Provide clear loading indicators for AI operations
56- Implement progressive disclosure for complex AI features
57- Give users control over AI behavior and settings
58- Provide feedback mechanisms for AI output quality