cloudflare

Use Workers for serverless edge functions, Pages for full-stack web apps with Git deploys, and Durable Objects for stateful coordination or real-time workloads. The Skill's decision trees route based on whether you need compute, state, or scheduled execution.

Use KV for high-read key-value data like config and sessions, D1 for relational SQLite queries, R2 for S3-compatible object storage, and Durable Object storage for strongly-consistent per-entity state. Queues handle async message processing.

Yes, Workers AI provides 50+ models including LLMs, embeddings, image generation, and speech-to-text via the env.AI binding. Vectorize adds vector search for RAG patterns, and AI Gateway provides caching, rate limiting, and analytics across providers.

The Skill explicitly instructs retrieval from live Cloudflare docs, the workers-types npm package, and the Wrangler config schema before citing limits, pricing, or API signatures. Reference files are starting points, not the source of truth.

The terraform reference covers provider v5 setup, API token authentication, resource configuration for zones, DNS, Workers, KV, R2, and D1, plus cf-terraforming for importing existing resources. It warns against mixing Terraform with wrangler.jsonc for the same resources.

Avoid Smart Placement for Workers serving static content, pure edge logic without backend calls, Workers without fetch handlers, and Pages assets with run_worker_first enabled. It only helps when backend latency dominates request duration.