@neon/ai-sdk-provider weekly downloads statistics

@neon/ai-sdk-provider

Community Vercel AI SDK provider for the Neon AI Gateway. Supports AI SDK v6 and v7 (ai@^6 or ai@^7).

The Neon AI Gateway is branch-scoped: each Neon project branch gets its own gateway host, and a platform token authorizes requests for that branch. Use the same neon(modelId) API across the branch's text models; the provider selects Anthropic Messages, OpenAI Responses, or Chat Completions for each model. Embedding models use neon.embeddingModel(modelId).

Use canonical model ids such as gpt-5-mini, llama-4-maverick, and gemini-3-flash, matching neon.com/models. The typed catalog includes the known model ids, and arbitrary strings are accepted so newly available models work before the types update.

Install

npm install @neon/ai-sdk-provider ai

Requirements: Node.js >= 20.19 with AI SDK 6. AI SDK 7 itself requires Node.js >= 22.

Configuration

The gateway URL is branch-scoped, so both values come from the Neon Console (your project → a branch → AI Gateway tab), or from neon env pull / neon dev:

NEON_AI_GATEWAY_BASE_URL="https://<branch-id>-api.ai.<region>.aws.neon.tech"
NEON_AI_GATEWAY_TOKEN="nt_live_..."

Usage

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

// Reads NEON_AI_GATEWAY_BASE_URL + NEON_AI_GATEWAY_TOKEN from the environment.
const { text } = await generateText({
  model: neon("gpt-5-mini"), // or "llama-4-maverick", "gemini-3-flash", …
  prompt: "Summarize Postgres for me.",
});

Or configure explicitly with createNeon:

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

const neon = createNeon({
  baseURL: process.env.NEON_AI_GATEWAY_BASE_URL,
  apiKey: process.env.NEON_AI_GATEWAY_TOKEN,
});

Routing

Model family	Endpoint	Why
Anthropic (claude-*)	Messages API	Anthropic tools, structured output, and reasoning
OpenAI (gpt-*, *-codex)	Responses API	Codex, reasoning, and built-in tools such as image generation
Everything else (Gemini, Llama, Qwen, gpt-oss, ...)	Chat Completions API	one streaming interface across the remaining model families

Routing matches on the model id.

Capabilities

generateText and streamText work with any text model available to your branch. generateObject, streamObject, and single- or multi-step tool calls work with OpenAI (including Codex), Meta, Alibaba, Zhipu AI, and Thinking Machines models. Gemini currently supports generateText and streamText; structured output and multi-step tools are not supported. Vision input works on models that accept images.

Claude models use the Messages API. On both that route and Chat Completions the provider removes call options the gateway rejects and reports each one in result.warnings instead of failing the request — see Dropped call options. Unsupported Responses API storage options throw before a request is sent — see Errors.

Which ids your branch serves is account-specific during the beta; GET $NEON_AI_GATEWAY_BASE_URL/v1/models is the authoritative list. NeonChatModelId accepts any string, so an id this package has not caught up with still works — and, equally, an id the gateway has retired still type-checks and fails at the gateway instead.

Dropped call options

Upstream backends behind the gateway accept different subsets of the OpenAI-style parameters the AI SDK emits, and sending one an upstream rejects is a hard 400. The provider drops those before the request and records a warning in result.warnings, so a call succeeds instead of failing on a parameter you passed in good faith.

These rules apply on the Anthropic Messages and Chat Completions routes. The Responses route (every gpt-* id) relies on the upstream OpenAI model's own stripping instead, so its behaviour and its warning wording are the AI SDK's rather than described here — but getNeonModelCapabilities still answers for those ids, and the gpt-5 row below is what it reports rather than something this provider drops.

Models	Dropped
Claude 4.7 and newer (claude-opus-4-7, claude-opus-4-8, claude-opus-5, claude-sonnet-5, claude-fable-5, claude-fable-5-1)	temperature, topP
All Claude	frequencyPenalty, presencePenalty, seed, reasoningEffort, and topP when you set both temperature and topP (Anthropic accepts only one)
gpt-oss	frequencyPenalty, presencePenalty, seed, stopSequences
Qwen, Gemma	frequencyPenalty, presencePenalty, seed
glm-5-2, glm-5-3-flash, inkling, kimi-k3	frequencyPenalty, presencePenalty
Meta Llama	frequencyPenalty, presencePenalty, seed
gemini-3-5-flash-lite	frequencyPenalty, presencePenalty
gemini-3-6-flash	temperature, topP, frequencyPenalty, presencePenalty
Every other Gemini	nothing
gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-5-pro (reported only — Responses strips these upstream)	temperature, topP

temperature on a Claude 5 model therefore has no effect. That is the gateway's behaviour, not a provider choice; sending it returns does not support the temperature parameter. Steer those models with Anthropic's own effort control instead — note that this is effort, not the OpenAI-style reasoningEffort, which every Claude id drops:

await generateText({
  model: neon('claude-opus-5'),
  prompt: 'Plan a schema migration.',
  providerOptions: { anthropic: { effort: 'high' } }, // low | medium | high | xhigh | max
});

Query the rules directly rather than hardcoding them:

import { getNeonModelCapabilities } from '@neon/ai-sdk-provider';…