Jev is now available on the AI Gateway
The AI Gateway now serves TypeSafe's System One decision models, Jev 1.14 and the open-weights OpenJev. Instead of generating text, they answer questions about your input with a choice, a score, or a probability that your code can act on directly. You're billed for input tokens only.
import { neon } from '@neon/ai-sdk-provider';
import { experimental_decide as decide } from 'ai';
const { answers } = await decide({
model: neon.decisionModel('typesafeai-jev-1-14-0'),
state: "We're growing fast and want to add more projects and team members.",
questions: {
upgrade: { type: 'boolean', instructions: 'Is the user interested in upgrading their plan?' },
},
});
console.log(answers.upgrade.probability); // 0.91See Decisions to get started. To see Jev in action, build a content moderation pipeline or use it to rerank hybrid search results with Lakebase Search.
Automatic top-up for AI Gateway credits
Keep your AI features running without watching your balance. Organization admins can now turn on auto-reload for AI Gateway credits: when your balance drops below a threshold you set, Neon automatically buys the amount you choose with your payment card on file. Set it up from the AI gateway credits card on the Billing page, or while you're adding credits. See Automatic top-up.

OAuth provider for Managed Better Auth (beta)
Managed Better Auth can now act as an OAuth 2.1 authorization server, so your users can sign in to your MCP server, API, or other apps with their existing accounts. To protect an MCP server, use the @neon/oauth SDK (beta):
import express from 'express';
import { createMcpAuth } from '@neon/oauth/mcp';
import { expressMcp } from '@neon/oauth/mcp/express';
const auth = createMcpAuth({
baseUrl: process.env.NEON_AUTH_BASE_URL!, // your branch's Auth URL
resource: 'https://mcp.example.com/mcp', // the MCP server URL set on the provider
});
const app = express();
app.use(expressMcp(auth)); // verifies access tokens and serves the resource metadataMCP clients identify themselves with Client ID Metadata Documents by default. Dynamic client registration is now a legacy option and is off by default. See OAuth Provider.
Scaffold functions and search the docs from the Neon CLI
neon functions new (Neon CLI 8.11 or later) scaffolds a Neon Function from a template. There are only a couple of starter templates so far, and we'll add more over time. Run neon functions templates list to see them. The quickest way to try it is neon functions new basic, which writes a small handler that returns JSON to functions/basic/index.ts and adds the function to neon.ts. Run it locally with neon dev, then deploy with neon config apply:
neon functions templates list
neon functions new basic
npm install @neon/config @neon/env
neon dev
neon config applyHave a function others could use? Templates live in the Neon Function Registry on GitHub, and anyone can propose one by opening a pull request. See the contributing guide.
neon docs search (Neon CLI 8.10 or later) searches neon.com from your terminal: the docs, API reference, Postgres tutorial, guides, FAQs, changelog, and blog. You don't need to be logged in. Use --category and --limit to narrow the results:
$ neon docs search "reset a branch" --limit 2
Manage branches › Reset a branch from parent
https://neon.com/docs/manage/branches
Branching with the Neon CLI › Resetting a branch from its parent
https://neon.com/docs/guides/branching-neon-cliA new support center in the Console
The Neon Console now has one place to get help. Open your organization menu and select Help > Get help to ask the AI assistant, open a support ticket, or reach the community. The options you see depend on your plan. See Support.

Fixes & improvements
Postgres extensions
plv8is deprecated on Neon, and now so areplcoffee(CoffeeScript) andplls(LiveScript), which ship with it on Postgres 14, 15, and 16.CREATE EXTENSION plcoffeeandCREATE EXTENSION pllsnow fail with a deprecation message, but existing installs keep working. If you still use them, migrate functions toplpgsqlor application code and drop the extensions. See The plv8 extension and Supported Postgres extensions.
Neon Functions
- The total size of a function's environment variables is now limited to 32 KiB, and deploys over the limit fail with an error. Previously, a deploy with 32 to 64 KiB of variables succeeded, but every request to the function returned a
404.
Object Storage
- Objects uploaded with multipart upload can now be up to 10 GiB, up from 5 GiB. Single-request uploads are still limited to 5 GiB. See Limits.
AI Gateway
- Fixed an issue where Gemini models on the chat completions endpoint returned
400 Unknown name "stream_options"when a request setstream_options, as the OpenAI SDK does to request usage in a stream. Requests withparallel_tool_calls: truenow work too. Gemini doesn't supportparallel_tool_calls: false, so leave that field out.
Realtime
- The Neon Realtime client now supports React Native (
@neon/realtime0.1.2 or later). Use the newwebSocketFactoryoption to customize the WebSocket connection, for example to send an allowed origin.
Neon Console
- We updated the Drizzle Studio integration behind the Tables page, where you browse and edit your data, to version 1.6.1. It fixes an issue with renaming schemas. For details, see the Neon Drizzle Studio Integration Changelog.
- Analyze is now Explain (Analyze) in the SQL Editor, with a tooltip warning that it runs data-modifying statements such as
INSERT,UPDATE,DELETE, andMERGE. Explain plans now include Neon'sPREFETCHandFILECACHEoptions, so Explain (Analyze) shows compute cache hits and prefetch statistics.
Neon CLI
- You can now create Claimable Neon projects in any AWS region that accepts new Neon projects. Pass
--region-idtoneon claim createorneon init --claimable(Neon CLI 8.8.0 or later), orregion_idin the API request. If you leave it out, the project lands inaws-us-east-2, as before. - Fixed
neon login --keyringfailing on Windows after the browser step.
Neon API
-
New
GET /projects/countendpoint (beta) returns the number of active projects in an organization, so you don't have to page throughGET /projects. With a personal API key, passorg_id. With an org API key, it's inferred.curl 'https://console.neon.tech/api/v2/projects/count?org_id=org-example-12345678' \ -H 'Accept: application/json' \ -H "Authorization: Bearer $NEON_API_KEY"{ "count": 12 }
Neon MCP Server
- The MCP server's documentation tools are now
search_docsandget_doc, replacinglist_docs_resourcesandget_doc_resource.search_docsreturns ranked results from across neon.com (the docs, API reference, guides, Postgres tutorial, FAQs, blog, and changelog), so your agent finds the right page without loading the whole docs index. If your agent's prompts or tool allowlists name the old tools, update them. See Neon MCP Server.
Vercel integrations
- With the Neon-Managed Vercel integration, you can now choose which Neon branch supplies Vercel Production credentials when you connect. To change it later, go to Integrations, click Manage next to Vercel, and open the Branches tab.
- If you installed Neon from the Vercel Marketplace, you can now choose which Neon branch backs Vercel Production. In the Neon Console, go to Integrations and click Settings next to Vercel. See Choose a production branch.
- New databases created from the Vercel Marketplace can turn on Neon Realtime (beta) at install.
- Fixed an issue where a failed update to a preview branch could stop the Neon-Managed Vercel integration from syncing new credentials to your Production environment, for example after a password rotation.
OpenTelemetry integration
- Fixed an issue where you couldn't save an OpenTelemetry integration for Dynatrace or Splunk Observability because credential validation failed.
- When you change an OpenTelemetry integration's endpoint, protocol, or endpoint overrides, you'll need to re-enter its credentials.








