Netlify’s AI Gateway is now generally available, and it could remove one of the most annoying steps in building an AI app: creating and wiring up a separate account for every model provider you want to test.
The launch matters because AI agents are getting better at generating entire apps, but they still tend to stop when the infrastructure work begins. Netlify’s pitch is simple: let the agent build the app, deploy it, and call supported models through the same platform.
What AI Gateway actually does
Netlify says AI Gateway is generally available to users on credit-based plans, with model access built into the platform. The service can call OpenAI, Anthropic, and Gemini models, while the documentation also lists OpenRouter support.
The practical change is that you don’t need to open an account with each provider, keep separate balances, or copy each provider’s API key into your Netlify project, according to Netlify’s documentation.
Netlify automatically supplies the environment variables and base URLs that official client libraries expect in its compute environments. That means a server-side function can use a provider SDK without the usual API-key setup, although OpenRouter needs its base URL passed explicitly.
The gateway is a proxy, not a new foundation model. Your request goes through Netlify to the provider that hosts the model, and Netlify converts the token usage into credits on your existing account. The model still comes from OpenAI, Anthropic, Google, or an OpenRouter provider.
That distinction matters for builders. The gateway can simplify operations, but it doesn’t make the underlying model cheaper, smarter, or more reliable. It gives you one place to manage access and usage while keeping the provider choice flexible.
The free plan has 300 credits
Netlify’s launch post says AI Gateway is available on all credit-based plans, and free plans include 300 credits across the platform, including AI inference.
Netlify’s documentation says $1 of AI model usage converts to 180 credits. The same page lists per-minute limits of 90 credits for Free, 450 for Personal, 1,800 for Pro, and 9,000 for Enterprise, so a busy public app still needs cost planning.
The service also has guardrails that are easy to overlook. Netlify describes rate limiting and credit controls as protection against runaway costs, and it recommends adding visitor-level rate limits to functions that use the gateway.
That’s useful for prototypes, but I wouldn’t treat the included credits as a promise of unlimited experimentation. Model calls can consume credits quickly, especially when an agent sends long prompts, retries failed work, or lets visitors trigger requests without a per-user limit.
There is a production-deploy catch
The quickstart walkthrough for the gateway walks through a Vite and React app, a Netlify Function, and the OpenAI client library. The example deploys the project to production before adding the AI function.
That order isn’t just tutorial housekeeping. Netlify says a project needs at least one production deployment before AI Gateway activates, which means a purely local experiment doesn’t get the full service by default.
The quickstart still looks refreshingly short. Install the Netlify CLI, initialize a project, deploy it, install the OpenAI SDK, and call the model from a server-side function. The example never asks you to create an OpenAI account or paste an OpenAI key into the project.
For teams, the one-vendor setup may be more valuable than the first demo. Billing appears on the Netlify bill, credentials can be managed centrally, and developers can test supported providers without creating a new account every time they compare models.
What I’d watch before using it everywhere
The convenience comes with platform dependence. If your app relies on Netlify’s injected variables, proxy behavior, credit accounting, and rate limits, moving the workload later means revisiting those assumptions. Netlify does let you provide your own API keys, but that partly gives back the simplicity AI Gateway is designed to provide.
The documentation says AI Gateway doesn’t store prompts or model outputs, and it describes zero-data-retention requirements for models served through the gateway. Those are useful assurances, but teams should still read the current security and privacy documentation before sending sensitive production data.
There are also technical limits. Netlify documents a 200,000-token context limit, no batch inference, no priority processing, and a requirement for a previous production deployment. Those constraints won’t matter for a small chatbot, but they can affect background jobs and evaluation pipelines.
If you’re comparing how different agent stacks handle deployment, Tony’s best AI coding agents guide and Hermes Agent release coverage give useful context for where this kind of platform layer fits.
My read is that AI Gateway is most interesting as agent infrastructure rather than as another model wrapper. The less time an agent spends asking a developer to create accounts, copy secrets, and reconcile invoices, the closer a request to build an app gets to meaning a working app.
If you’re already deploying on Netlify, I’d test one small server-side feature first. Track credits, inspect the generated requests, add visitor-level limits, and compare the same call with your own provider key. That will tell you whether the operational convenience is worth making Netlify part of the app’s model-access layer.
