Netlify has shipped a point release into its AI Gateway, and the changelog is short enough to quote nearly whole. TypeSafe Jev 1.14.0 Now Available in AI Gateway says the model “is now available through Netlify’s AI Gateway with zero configuration required,” installed through @typesafe-ai/sdk and called from a Netlify Function “without managing API keys or authentication.” The gateway holds the credentials and usage is billed to Netlify credits. The example pins jev-1.14.0 by name and asks a single typed question, a choice that labels a support ticket as a bug, a feature request, or a question, and the release extends the model to Scheduled, Background, and Edge Functions.
There is no vendor figure to discount this time. The October changelog that introduced Jev carried TypeSafe’s latency range and token budget; this one carries a version number and a list of function types. What it does claim, that callers get “automatic access to Netlify’s caching, rate limiting, and authentication infrastructure,” is a platform claim, and the interesting part is that pinning an explicit version is now the documented pattern. When we wrote up the first Jev changelog the SDK defaulted to a moving jev-latest alias. A named version in the canonical example is the more defensible way to put a classifier on a request path.
The catalog has moved since October too. The Netlify provider page lists 41 API pages, and one of them is new to this story: the Netlify AI Gateway API, the surface the changelog is about, which had no page of its own the last time we looked. The code in the example runs where the Netlify Function API and the Netlify Edge Function API manage it, and the credentials the gateway spares you sit beside the Netlify Environment Variables API you would otherwise have used to store them. The agentic access profile maps 133 operations, 73 acting and 1 human-in-the-loop.
The Kin Score is 72.8, exemplar band, up from 71.9, with access clarity at 86.8, operational transparency at 84.2, and developer ergonomics at 79.8 carrying it. Contract quality rose to 55.5 from 51.9; contract governance is unchanged at 31.8. Agent Readiness went the other way, from 60.0 to 58.8, still agent-ready, with the MCP server, agent skills, the well-known catalog, dry-run mode, and dynamic client registration lit, and idempotency, consent identity, the agent card, agentic commerce, and pagination ratio dark. That is the quiet finding under a two-paragraph changelog: the provider got easier to read as a contract in the same month it got slightly harder for an agent to use, and a single score would have shown only the first half.