KV, Vectorize & AI

in-progress

Three smaller bindings grouped together — KV (eventually-consistent edge cache), Vectorize (vector search), AI Gateway (proxied LLM calls with caching and analytics).

Source reference: cli/internal/serverless/cloudflare_{kv,vectorize,ai}.go

KV: when to use

TODO eventually-consistent global key/value, low write rate, high read rate. Examples — feature flags, configuration, session lookup.

KV: declaring & reading

TODO YAML shape (namespace name + binding), env.MY_KV.get/put/list, TTL semantics, eventual consistency caveats.

Vectorize: when to use

TODO semantic search, RAG retrieval, similarity over embeddings. NOT a relational store.

Vectorize: declaring an index

TODO YAML shape — index name, dimensions, metric (cosine / dot / euclidean), preset embedding model (or BYO).

Vectorize: writing & querying

TODO code example — env.INDEX.upsert([...]), query(vector, { topK, filter }). Pairing with Workers AI for embeddings.

AI Gateway: what it gives you

TODO central proxy for OpenAI / Anthropic / etc. — caching, rate limit, logging, fallback routing. NextDeploy wires the binding.

AI Gateway: declaring

TODO YAML shape — gateway slug, default provider, cache TTL, retry config.

AI Gateway: using from code

TODO code example — env.AI_GATEWAY.run(...) or fetch through the gateway URL exposed via binding.

Pricing summary

TODO KV (per read / write / storage), Vectorize (per dim-vector stored + queries), AI Gateway (free; underlying model usage billed by provider).