Skip to content

Limitations & Considerations

Consistency

  • Eventual consistency: Reads may lag behind writes
  • Vectorize constraint: Built on eventually consistent infrastructure
  • Queue dependencies: Processing delays affect freshness

Storage Limits

  • D1: 10GB database limit
  • Vectorize: Dimension and metric constraints
  • KV: 1GB per namespace

Development Constraints

  • Cloudflare lock-in: Tight ecosystem coupling
  • Context generation: Manual developer input required
  • Learning curve: New retrieval model concepts

Operational Considerations

  • Temporal decay: Requires periodic cleanup jobs
  • Queue management: PMQ vs Cloudflare Queues trade-offs
  • Memory growth: Unbounded memory accumulation

Performance Trade-offs

  • SQL scoring: Scalable but limited complexity
  • Over-fetching: Semantic search retrieves extra for filtering controlable through topK environment variable.
  • Edge compute: Low latency but worker constraints