/create-vector-index

Arguments: $ARGUMENTS

Goal: Create a Redis vector index configured for similarity search with support for FLAT or HNSW algorithms, custom distance metrics, and metadata fields.

Core Principles:

• Detect existing Redis configuration before creating index
• Validate schema and parameters before execution
• Support both FLAT (exact search) and HNSW (approximate) algorithms
• Configure appropriate distance metrics (COSINE, L2, IP)
• Add metadata fields for filtering and hybrid search

Phase 1: Discovery Goal: Understand project context and requirements

Actions:

• Parse $ARGUMENTS to extract:
  • Index name (required)
  • Index type: --type=FLAT or --type=HNSW (default: HNSW)
  • Distance metric: --metric=COSINE|L2|IP (default: COSINE)
  • Vector dimensions: --dimensions=N (required)
  • Additional metadata fields (optional)
• Detect project structure and Redis configuration:
  • Check for existing Redis connection files
  • Look for environment variables or config files
  • Example: !{bash find . -name ".env" -o -name "redisconfig*" 2>/dev/null | head -5}
• If any required parameters are missing, use AskUserQuestion:
  • What is the vector dimension? (e.g., 1536 for OpenAI embeddings)
  • Which index type? (FLAT for exact search, HNSW for approximate)
  • Which distance metric? (COSINE for normalized vectors, L2/IP for others)
  • What metadata fields to include? (e.g., user_id, category, timestamp)

Phase 2: Validation Goal: Verify inputs and environment are ready

Actions:

• Validate index parameters:
  • Index name is valid (alphanumeric, underscores)
  • Dimensions is positive integer
  • Type is either FLAT or HNSW
  • Metric is COSINE, L2, or IP
• Check Redis connection configuration:
  • Verify REDIS_URL or connection settings exist
  • Example: @.env (if exists)
• Confirm prerequisites:
  • Redis Stack or RedisJSON module available
  • Vector search capability enabled

Phase 3: Implementation Goal: Create vector index using specialized agent

Actions:

Task(description="Create vector index", subagent_type="vector-index-architect", prompt="You are the vector-index-architect agent. Create a Redis vector index for $ARGUMENTS.

Context:

• Project directory detected in Phase 1
• Index parameters validated in Phase 2
• Redis configuration available

Requirements:

• Create FT.CREATE command with proper schema
• Configure index type (FLAT or HNSW) based on parameters
• Set up vector field with correct dimensions and distance metric
• Add metadata fields for filtering (if specified)
• Include schema validation
• Generate example queries for the index
• Provide connection code snippet

Index Configuration:

• Name: [extracted from $ARGUMENTS]
• Type: [FLAT or HNSW]
• Metric: [COSINE, L2, or IP]
• Dimensions: [extracted from $ARGUMENTS]
• Metadata: [extracted from $ARGUMENTS or defaults]

Expected output:

1. Redis FT.CREATE command for index creation
2. Schema validation script
3. Example search queries (KNN, range, hybrid)
4. Connection code snippet for application integration
5. Performance tuning recommendations")

Phase 4: Verification Goal: Validate the created index configuration

Actions:

• Review agent's output for completeness
• Check that FT.CREATE command includes:
  • Correct index type (FLAT or HNSW)
  • Vector field with proper VECTOR parameters
  • Distance metric configuration
  • All metadata fields
• Verify example queries are syntactically correct
• Confirm connection code matches project language/framework

Phase 5: Summary Goal: Report index creation and next steps

Actions:

• Display index configuration summary:
  • Index name and type
  • Vector dimensions and metric
  • Metadata fields included
  • Algorithm parameters (if HNSW: M, EF_CONSTRUCTION, EF_RUNTIME)
• Show FT.CREATE command to execute
• Provide example search queries
• Suggest next steps:
  • Execute FT.CREATE command on Redis instance
  • Insert sample vectors for testing
  • Integrate search queries into application
  • Monitor index performance and adjust parameters