Neon AI Gateway

FIRST: Use the parent neon skill for a Neon overview, getting started with Neon, Neon development best practices, and more.

If the neon skill is not installed, fetch it from https://neon.com/docs/ai/skills/neon/SKILL.md (opens in a new tab) or install it with:

neon skills -s neon -y

Neon AI Gateway

Currently available in aws-us-east-2, aws-us-east-1, aws-eu-central-1, and aws-ap-southeast-1.

The Neon AI Gateway is the LLM inference layer built into your Neon branch: one API and one Neon credential give you access to frontier and open-source models from many providers (Anthropic, OpenAI, Google, Meta, and more), all hosted and powered by Databricks. The catalog shifts over time, so treat /v1/models and the models.dev Neon page (opens in a new tab) as the source of truth rather than a fixed provider list. Your existing OpenAI/Anthropic/Gemini SDK works by changing only the base URL.

Use this skill to help the user send model calls through the gateway, wire it into the AI SDK or Mastra, and switch providers without rewiring code. Deliver a working inference request, a configured agent, or a precise answer from the official Neon docs.

When to Use

Reach for the AI Gateway whenever an app or agent needs to call an LLM and the user would rather not manage model providers themselves:

• One credential instead of many provider accounts. A single Neon credential reaches the entire model catalog across every provider Databricks hosts. No separate OpenAI / Anthropic / Google billing, keys, or signups to provision and rotate.
• Switch models without rewiring. The unified endpoint is OpenAI-compatible and works with every model in the catalog — change one model field to move between Claude, GPT, and Gemini. Standard SDKs (OpenAI, Anthropic, google-genai) work with just a base-URL change.
• AI follows your branches. Each branch has its own gateway endpoint, scoped with the same lineage as your database. AI requests from a preview/feature branch are isolated to that branch — the same isolation your data already gets — which makes preview, CI, and agent environments self-contained.
• No extra infrastructure, and it’s already next to your data. The gateway lives inside your Neon project (and is injected into Neon Functions automatically), runs on the same Databricks infrastructure that serves trillions of tokens a month, and supports streaming (SSE) out of the box.

If the user already has a deep, single-provider integration and no interest in Neon branching or multi-model routing, a direct provider SDK is fine — but the moment they want one credential, model portability, or branch-scoped AI, this is the reason to use it.

What It Does

• One API for all models — Frontier and open-source models behind a single endpoint, addressed by their catalog ID (e.g. claude-sonnet-4-6, gpt-5-mini, gemini-3-flash).
• Standard SDKs, one URL change — OpenAI SDK and AI SDK (OpenAI-compatible MLflow/Responses routes), Anthropic SDK (native Messages), google-genai (native Gemini).
• Branch-scoped — Each branch gets its own gateway host; the Neon credential authorizes requests for that branch and its descendants.
• Streaming — Server-sent events work on all endpoints with no extra configuration.

Availability

Check these preconditions before setting anything up:

The AI Gateway is currently available in aws-us-east-2, aws-us-east-1, aws-eu-central-1, and aws-ap-southeast-1. Foundation model access requires a paid Neon plan. Confirm the user’s project is in one of these regions.

Enabling the gateway: plan and model-catalog gating

The AI Gateway is credential-gated rather than a provisioning step, but two plan limits gate it — one blocks provisioning, the other only trims the catalog — and the CLI surfaces each:

• Free plan → provisioning is blocked. neon config apply / deploy and neon checkout refuse to enable the gateway on a Free plan (the gateway can’t serve requests there), with a friendly “upgrade to a paid plan, or remove aiGateway“ error. A dry-run neon config plan and neon env pull don’t provision, so they only warn. So: to use the gateway the project’s account must be on a paid Neon plan.
• Paid plan with a reduced model catalog. On a paid plan the gateway provisions and serves, but an account can start with a trimmed catalog — some flagship models (e.g. Anthropic Opus, OpenAI Codex / *-pro) are missing from GET /v1/models. This is expected; neon env pull (and the env pull bundled into apply / deploy / checkout) warns and links the user to their branch’s AI Gateway page in the Neon Console (https://console.neon.tech/app/projects/<project-id>/branches/<branch-id>/ai-gateway) to request access to more models. Verify what’s actually available for the branch by reading /v1/models (see the models section below) rather than assuming the full catalog.

When helping a user debug “the gateway isn’t working” or “a model is missing”, use /v1/models plus the account’s plan to distinguish these two cases — a Free plan blocks provisioning entirely, while a reduced catalog on a paid plan just needs a model-access request.

Setup

The gateway is part of neon.ts (see the neon skill for the branch-first workflow and neon.ts basics). Enable it with aiGateway:

// neon.ts
import { defineConfig } from "@neon/config/v1";

export default defineConfig({
  aiGateway: true,
});

neon deploy   # provisions the gateway on the linked branch

Neon Infrastructure as Code (neon.ts)

The aiGateway toggle above is part of neon.ts, Neon’s infrastructure-as-code file — one TypeScript file declares the gateway alongside every other branch service, in version control (see the neon skill for the full reference). Reconcile it against a branch the Terraform way:

neon config status   # print the branch's live config (is the gateway on?)
neon config plan     # dry-run diff of what apply would change
neon config apply    # enable the gateway on the branch  (neon deploy is an alias)

The gateway is branch-scoped: each branch gets its own gateway host. When a neon.ts is present, neon checkout applies the policy as it creates a branch, so a fresh preview/CI branch comes up with the gateway already enabled. Checking out an existing branch doesn’t reconcile it — run neon deploy to apply changes. Provisioning (config apply / deploy), link, and checkout also pull the branch’s gateway credentials into your local .env.local, so local runs hit the same branch gateway as the deployed function (no manual env pull needed).

Environment Variables

When aiGateway is enabled, Neon injects the gateway credentials as Neon-branded env vars. Inside a deployed Neon Function these are injected automatically; locally, neon env pull writes them to .env/.env.local (or use neon-env run -- <cmd> to inject at runtime without a file):

Neon injects only these two vars — it does not set OPENAI_API_KEY / OPENAI_BASE_URL. The @neon/ai-sdk-provider and Mastra’s neon/<model> read NEON_AI_GATEWAY_* directly (zero config); for the plain OpenAI SDK / @ai-sdk/openai, build the client’s apiKey + baseURL from them (shown below), or set your own OPENAI_* by hand (env pull leaves user-set vars untouched).

NEON_AI_GATEWAY_BASE_URL is the bare host — you append the dialect path yourself (which is exactly what the @neon/ai-sdk-provider does for you). The routes under the host are:

• /v1 — unified, OpenAI Chat Completions-compatible; recommended default, works with every provider (/v1/chat/completions).
• /openai/v1 — OpenAI Responses API (required for gpt-5-…-codex variants and gpt-5-5-pro); the @ai-sdk/openai provider uses the Responses API by default (/openai/v1/responses).
• /anthropic — native Anthropic Messages (extended thinking, prompt caching). Give the Anthropic SDK this as its base URL and it appends /v1/messages itself, so the full request path is /anthropic/v1/messages.
• /gemini — native Gemini generateContent. Give google-genai this as its base URL and it appends /v1beta/models/<model>:generateContent itself, so the full request path is /gemini/v1beta/models/<model>:generateContent.

So ${NEON_AI_GATEWAY_BASE_URL}/v1 is the chat-completions endpoint and ${NEON_AI_GATEWAY_BASE_URL}/openai/v1 the OpenAI Responses endpoint (both appended by you); for the native Anthropic and Gemini dialects you hand the SDK the shorter /anthropic or /gemini base and it appends the rest. See Use with Plain SDKs below.

For typed, validated access to the injected credentials, pass the same neon.ts config object to parseEnv from @neon/env — it returns an env.aiGateway namespace (apiKey, baseUrl) derived from your config.

Build Agents with the Vercel AI SDK (Recommended)

The Vercel AI SDK (opens in a new tab) is the recommended way to call the gateway and build agents from TypeScript: one set of primitives (generateText, streamText, tool calling, structured output) over every catalog model, with first-class streaming for the long agent responses Neon Functions are built to host.

The dedicated @neon/ai-sdk-provider reads NEON_AI_GATEWAY_BASE_URL + NEON_AI_GATEWAY_TOKEN from the injected env with zero config and routes each model to the best endpoint (Anthropic → Messages, OpenAI/Codex → Responses, everything else → MLflow). On a Neon Function that streams text and generates images, just pick a catalog model:

import { neon } from "@neon/ai-sdk-provider";
import { streamText } from "ai";

const result = streamText({
  model: neon("gpt-5-mini"), // or claude-sonnet-4-6, gemini-3-flash, ...
  messages,
  tools: {
    image_generation: neon.tools.imageGeneration({
      outputFormat: "jpeg",
      size: 

A single completion is the same provider with generateText:

import { neon } from "@neon/ai-sdk-provider";
import { generateText } from "ai";

const { text } = await generateText({
  model: neon("claude-haiku-4-5"), // or gpt-5-3-codex, gemini-3-flash, ...
  prompt: "Summarize Postgres for me.",
});

Prefer @neon/ai-sdk-provider over the bare @ai-sdk/openai openai(): Neon injects only NEON_AI_GATEWAY_*, not OPENAI_*, so openai() won’t pick up the gateway from the env on its own. If you do use @ai-sdk/openai, configure it explicitly with createOpenAI({ apiKey: process.env.NEON_AI_GATEWAY_TOKEN, baseURL: ${process.env.NEON_AI_GATEWAY_BASE_URL}/openai/v1 }).

To build an agent — a model that calls tools in a loop and then answers — add tools and a stopWhen budget. The loop runs in-process, so on a Neon Function it isn’t cut off by lambda-style timeouts:

import { neon } from "@neon/ai-sdk-provider";
import { generateText, tool, stepCountIs } from "ai";
import { z } from "zod";

const { text } = await generateText({
  model: neon("claude-sonnet-4-6"),
  prompt: "How many open todos do I have, and what's the oldest one?"

For a full AI SDK agent deployed as a Neon Function (streaming, tool calling, image generation, persistence), see the neon-functions skill’s references/ai-sdk.md (opens in a new tab).

Build Agents with Mastra (Recommended)

Mastra (opens in a new tab) is the recommended framework when you want batteries-included agents — built-in memory, tools, workflows, and tracing — with the model still pointed at the gateway. With @mastra/core 1.47+, use a neon/<model> magic string; Mastra reads NEON_AI_GATEWAY_BASE_URL and NEON_AI_GATEWAY_TOKEN from the environment (injected by neon deploy when aiGateway is enabled). Use parseEnv only for other declared services (e.g. env.postgres.databaseUrl for @mastra/pg memory):

import { Agent } from "@mastra/core/agent";
import { parseEnv } from "@neon/env";
import config from "../neon";

const env = parseEnv(config);

export const personalAssistant = new Agent({
  id: "personal-assistant",
  name: 

Use with Plain SDKs (Lower-Level)

When you don’t need an agent framework — a single completion, an existing provider-SDK integration, or native provider features — call the gateway with the plain SDKs. Neon injects the NEON_AI_GATEWAY_* vars (not OPENAI_*), so set the client’s apiKey + baseURL from them. For the OpenAI Responses dialect (/openai/v1):

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.NEON_AI_GATEWAY_TOKEN,
  baseURL: `${process.env.NEON_AI_GATEWAY_BASE_URL}/openai/v1`,
});

const res = await client.responses.create({
  model: "gpt-5-mini"

For the unified chat-completions dialect, point baseURL at /v1 instead:

const client = new OpenAI({
  apiKey: process.env.NEON_AI_GATEWAY_TOKEN,
  baseURL: `${process.env.NEON_AI_GATEWAY_BASE_URL}/v1`,
});

const res = await client.chat.completions.create({
  model: "claude-sonnet-4-6",
  messages: [{ role: "user", content: "What is Neon?"

The Anthropic SDK and google-genai work the same way for native provider features — point the Anthropic SDK at ${NEON_AI_GATEWAY_BASE_URL}/anthropic (it appends /v1/messages itself) and google-genai at ${NEON_AI_GATEWAY_BASE_URL}/gemini (it appends /v1beta/models/...).

Model Identifiers

Use a model’s catalog ID directly in the model field — e.g. claude-sonnet-4-6, gpt-5-mini, gemini-3-flash. No provider prefix is needed. To look up the exact identifiers the gateway serves, which underlying model each maps to, and their context windows, pricing, and capabilities, use any of:

• models.dev Neon provider page: https://models.dev/providers/neon (opens in a new tab) — the canonical, always-current list of the Neon provider’s model IDs and their underlying models. The machine-readable catalog is at https://models.dev/api.json (opens in a new tab) (the neon key).
• Models doc: see Further Reading.

List Available Models at Runtime (/v1/models)

The gateway also exposes the model catalog live from your own branch endpoint, so an app or agent can discover exactly which models this branch serves without hard-coding the list. It is an OpenAI-compatible list endpoint, served only on the unified dialect (/v1):

curl "$NEON_AI_GATEWAY_BASE_URL/v1/models" \
  -H "Authorization: Bearer $NEON_AI_GATEWAY_TOKEN"

• GET ${NEON_AI_GATEWAY_BASE_URL}/v1/models → 200
• GET ${NEON_AI_GATEWAY_BASE_URL}/openai/v1/models → 404 (not served on the Responses dialect — use /v1)

Getting the credentials for the request. Both values come from the same branch-scoped Neon credential the gateway uses everywhere else — you never manage a provider key:

• Provision via neon.ts (recommended). Enable aiGateway in neon.ts and run neon deploy (or neon config apply). Provisioning, neon link, and neon checkout pull NEON_AI_GATEWAY_TOKEN + NEON_AI_GATEWAY_BASE_URL into your local .env.local; inside a deployed Neon Function they’re injected automatically. See Setup and Environment Variables above.
• Pull into the environment via CLI. neon env pull writes the two vars to .env/.env.local, or neon-env run -- <cmd> injects them at runtime without a file — but only when neon.ts declares aiGateway; the vars are never pulled off branch state alone.
• Provision via the Console UI. Enable the AI Gateway on the branch in the Neon Console and copy the branch’s gateway base URL and a Neon credential (token) from the project’s connection/credentials view.

Any Neon credential (nt_live_...) valid for the branch works as the bearer token; NEON_AI_GATEWAY_BASE_URL is the bare branch host (no path).

Response shape — OpenAI/OpenRouter-compatible list:

{
  "object": "list",
  "data": [
    {
      "id": "claude-sonnet-4-6",              // catalog model ID — use directly in the `model` field
      "canonical_slug": "claude-sonnet-4-6",
      "name": "Claude Sonnet 4.6",            // human-readable display name

Note: context_length, pricing, and per_request_limits are currently null and created is 0 for every entry — for context windows, pricing, and capabilities use the models.dev catalog above. Use /v1/models when you need the live, branch-scoped list of servable model IDs (e.g. to populate a model picker or validate a model before a request).

Neon Documentation

The Neon documentation is the source of truth and the AI Gateway is evolving rapidly, so always verify against the official docs. Any doc page can be fetched as markdown by appending .md to the URL or by requesting Accept: text/markdown. Find the right page from the docs index (https://neon.com/docs/llms.txt (opens in a new tab)) and the changelog announcements.

Further Reading

• https://neon.com/docs/ai-gateway/overview.md (opens in a new tab)
• https://neon.com/docs/ai-gateway/get-started.md (opens in a new tab)
• https://neon.com/docs/ai-gateway/models.md (opens in a new tab)
• https://neon.com/docs/ai-gateway/chat-completions.md (opens in a new tab)
• https://neon.com/docs/ai-gateway/anthropic-messages.md (opens in a new tab)
• https://neon.com/docs/ai-gateway/openai-responses.md