/add-rag

Arguments: $ARGUMENTS

Goal: Implement a complete RAG pipeline using Google Gemini File API as the primary RAG solution with Redis as an intelligent caching layer for Gemini responses, embeddings, and file processing results.

Core Principles:

• Google Gemini File API is the PRIMARY RAG solution (handles embeddings, file uploads, retrieval)
• Redis provides CACHING layer (semantic cache for responses, embedding cache, file processing cache)
• Optimize for cost/speed by caching expensive Gemini API calls
• Support optional LangChain/LlamaIndex integration for flexibility
• Never hardcode API keys - use placeholders and environment variables
• Detect existing project structure before implementing

Phase 1: Discovery Goal: Understand project context and RAG requirements

Actions:

• Parse $ARGUMENTS for feature description or use case
• If unclear, use AskUserQuestion to gather:
  • What documents/data will be processed? (PDFs, text files, web pages, etc.)
  • What queries will users ask? (search, Q&A, summarization, etc.)
  • Caching strategy preference? (aggressive for cost savings, conservative for freshness)
  • Integration preference? (Gemini only, or also LangChain/LlamaIndex)
• Detect project type and framework
• Example: !{bash ls package.json requirements.txt pyproject.toml 2>/dev/null}
• Check for existing Redis configuration
• Example: @.env.example

Phase 2: Architecture Planning Goal: Design RAG pipeline architecture

Actions:

• Load existing project configuration files for context
• Determine architecture based on requirements:
  • Gemini File API Layer: File uploads, vector embeddings, semantic retrieval
  • Redis Caching Layer: Semantic cache, embedding cache, file processing results
  • Optional Framework Integration: LangChain or LlamaIndex abstractions
• Identify required components:
  • File upload/processing pipeline
  • Embedding generation (via Gemini)
  • Vector similarity search (via Gemini)
  • Redis semantic caching configuration
  • Query processing and response generation
• Present architecture plan to user for confirmation

Phase 3: Implementation Goal: Build RAG pipeline with Gemini + Redis

Actions:

Task(description="Implement RAG pipeline with Google Gemini File API and Redis caching", subagent_type="redis:rag-specialist", prompt="You are the redis:rag-specialist agent. Implement a complete RAG pipeline with Google Gemini File API and Redis caching for $ARGUMENTS.

Context from Discovery:

• Feature description: $ARGUMENTS
• Project type detected
• Existing Redis configuration (if any)
• User requirements gathered

Architecture Requirements:

• PRIMARY: Google Gemini File API for RAG
  • File upload and processing (PDF, text, web)
  • Vector embedding generation
  • Semantic search and retrieval
  • Document chunking strategies
• CACHING: Redis for performance optimization
  • Semantic cache for Gemini API responses
  • Embedding cache to avoid recomputation
  • File processing results cache
  • Query result cache with TTL
• OPTIONAL: LangChain/LlamaIndex integration
  • Only if user requested framework integration
  • Use as abstraction layer over Gemini + Redis

Security Requirements:

• Use environment variables for API keys (GEMINI_API_KEY, REDIS_URL)
• Create .env.example with placeholders only
• Never hardcode credentials
• Document key acquisition in setup guide

Implementation Deliverables:

1. Gemini File API integration code
   • File upload handlers
   • Embedding generation functions
   • Semantic search/retrieval logic
2. Redis caching layer
   • Semantic cache implementation
   • Embedding cache with TTL
   • File processing cache
3. RAG pipeline orchestration
   • Query processing flow
   • Cache-first strategy
   • Fallback to Gemini API
4. Configuration files
   • .env.example with placeholders
   • Redis cache configuration
   • Gemini API settings
5. Documentation
   • Setup guide with API key instructions
   • Usage examples
   • Caching strategy explanation
   • Cost optimization tips
6. Example code
   • Sample RAG queries
   • File upload examples
   • Cache monitoring

Expected Output: Complete RAG implementation with Gemini as primary engine and Redis as intelligent cache")

Phase 4: Validation Goal: Verify RAG pipeline functionality

Actions:

• Check that all required files were created
• Verify Redis caching configuration
• Validate Gemini API integration setup
• Run example queries if test data provided
• Example: !{bash python -c "import redis; redis.Redis().ping()" 2>/dev/null && echo "Redis OK" || echo "Redis not running"}

Phase 5: Summary Goal: Document implementation and next steps

Actions:

• Summarize RAG pipeline architecture:
  • Gemini File API components implemented
  • Redis caching layers configured
  • Framework integrations (if any)
• List files created/modified
• Highlight key configuration:
  • Cache TTL settings
  • Embedding dimensions
  • Chunk sizes and strategies
• Provide next steps:
  • Set up Gemini API key (link to console)
  • Configure Redis connection
  • Upload first documents
  • Test sample queries
• Show cost optimization tips:
  • Cache hit rate monitoring
  • Embedding reuse strategies
  • Query result caching