> Full Neon documentation index: https://neon.com/docs/llms.txt
> IMPORTANT: If this page contains inaccurate or outdated information, report it: POST to https://neon.com/api/docs-feedback with {"feedback": "describe the issue", "path": "/changelog/2026-03-27"}

# Cache prewarm, Stripe Projects integration, Neon Auth webhooks with Resend, and more

## Neon Postgres in Stripe Projects

**Neon is now part of** [**Stripe Projects**](https://projects.dev). Stripe Projects is a **Stripe CLI** workflow for **hooking an app up to backends**. You pick services from a catalog (databases, hosting, auth, and more), provision them into **your** provider accounts, and get **connection strings and keys** in your environment so your **agent** can run against a live backend.

Example Stripe CLI flow (after you install the Stripe CLI):

```bash
stripe projects init my-app   # set up your project
stripe projects catalog        # browse available services
stripe projects add neon/postgres       # provision a Neon database
```

Learn more in [Neon works with Stripe Projects for agentic provisioning](https://neon.com/blog/neon-works-with-stripe-projects-for-agentic-provisioning).

## Automatic cache prewarming for compute updates

To apply [updates](https://neon.com/docs/manage/updates) to your Neon compute (Postgres upgrades, security patches, and the like), we restart the compute where Postgres runs during your [update window](https://neon.com/docs/manage/updates#updates-on-paid-plans). The restart itself typically takes only a few seconds, but in-memory caches are left cold, which can impact query performance until they warm up again.

To protect performance, we now prewarm your compute's cache during the update process without affecting restart times. There are no additional compute or storage costs associated with this enhancement.

For the technical details, see our blog post: [Zero-Downtime Patching Part 1: Prewarming](https://neon.com/blog/prewarming).

## Snapshots billed at $0.09 per GB-month in May

[Snapshots](https://neon.com/docs/guides/backup-restore) have been in Beta and available free to all users. Beginning May 1, 2026 Snapshot storage will be billed at $0.09/GB-month.

## Neon Auth webhooks walkthrough with Resend

We announced support for [Neon Auth webhooks](https://neon.com/docs/changelog/2026-03-13#neon-auth) a couple of weeks ago. With webhooks, your app receives auth events over HTTP so you can send OTPs and other auth messages through your own email, SMS, or WhatsApp providers, validate signups, or sync to other systems.

This week we have a new guide, [**Customizing Neon Auth with Webhooks**](https://neon.com/guides/neon-auth-webhooks-nextjs), that walks you through setting up a **Next.js** app that handles webhooks, sends OTP email with **Resend**, tests locally with **ngrok**, and uses a blocking handler to reject signups you don't want. You can reference this guide alongside our [Webhooks](https://neon.com/docs/auth/guides/webhooks) docs when you are ready to implement.

## AI agents and instant database branching

Neon's instant **database branching** and **AI coding agents** work well together. Branching gives each run an isolated Postgres database in seconds, so agents can perform schema migrations, run tests, and try risky changes safely. The new guides below demonstrate this workflow with Codex and Claude Code.

- [**Safe AI-powered schema refactoring with OpenAI Codex and Neon**](https://neon.com/guides/openai-codex-neon-mcp): Use **OpenAI Codex CLI** with the **Neon MCP Server** so Codex can create an isolated branch, run Drizzle migrations, and validate schema refactors safely. In your project, wiring Codex to Neon is a small MCP configuration:

  ```toml
  [mcp_servers.neon]
  url = "https://mcp.neon.tech/mcp"
  bearer_token_env_var = "NEON_API_KEY"
  ```

- [**Isolated Subagents: Running Claude Code in parallel with Neon Database Branching**](https://neon.com/guides/isolated-subagents-neon-branching): Use a `post-checkout` hook to give each **Claude Code subagent** its own Git worktree and **Neon branch**, so parallel agents can run without code or database collisions.