Neon Blog

@neon/sdk is a new TypeScript client for the Neon API, replacing @neondatabase/api-client. It is fetch-based, zero dependencies, ESM-only, and generated from the OpenAPI spec with an ergonomic layer that handles authentication, typed error results, automatic retries, readiness polling, pagination, and multi-step workflows like snapshot preview-before-commit and project transfer across organizations.

Neon has launched neon.ts, a TypeScript configuration file for managing Neon services as infrastructure as code. It allows developers to declare services like functions, buckets, and AI gateways, and provides type-safe environment variables for applications.

Neon blog announces that the Vercel CLI now allows agents to provision a Neon database for a project with the command vercel integration add neon. This integration keeps the entire flow in the terminal, reducing steps for both humans and AI agents.

Three new Neon CLI commands, neonctl link, checkout, and env pull, streamline branch-first development by connecting workspaces to Neon projects, managing development variables, and switching branches. These commands can be used directly or handed to coding agents, with options to inject environment variables at runtime or integrate with env managers.

Starting June 1, 2026, all Neon paid plans include 500 GB of monthly data transfer, up from 100 GB. The change is automatic with no action required.

Neon published Neon Slop Fork, a fully functional dashboard rebuild created by a coding agent using only the public Neon API, OpenAPI spec, and TypeScript SDK. The project demonstrates API-first design for agent accessibility and serves as a template for building custom Neon management surfaces.

Article describes how Neon's lakebase architecture handles cloud failures amid increased agent workloads and control-plane strain, and outlines lessons informing the reliability roadmap.

Neon disabled full-page writes (FPW), a Postgres durability feature made redundant by Neon's separated compute-storage architecture, achieving up to 5x write throughput improvement on large instances and 94% reduction in WAL traffic. Image generation was moved from the compute layer to Neon's distributed storage system, which now generates full-page images based on actual page changes rather than Postgres checkpoints, preserving read performance while eliminating WAL overhead.

Organizations can now set monthly spending limits in the Neon console; admins receive email alerts when spend reaches 80% and 100% of the limit, with a banner visible to all team members. Automatic project suspension on limit breach is coming soon.

Neon's documentation engineer shares lessons from making docs machine-readable for AI agents, including MDX-to-Markdown conversion pipelines, content negotiation via Accept headers, llms.txt indexing strategies, and findings from scanning 250+ tech documentation sites. Key patterns include serving Markdown by appending .md to URLs (53% of surveyed sites), honoring Accept: text/markdown (41%), and structuring llms.txt indexes with sections rather than flat lists.

An official Neon plugin is now available in the OpenAI Codex marketplace, connecting Codex directly to Neon databases through MCP. The plugin bundles tools to create and manage projects, branches, and databases; run SQL queries; and validate connections, plus skills for Neon-specific workflows like branching strategies, autoscaling, and egress optimization.

Neon now automatically prewarms compute caches before planned restarts, eliminating the performance degradation that typically follows database restarts. The system spins up a new compute node in the background, loads cached pages from shared storage, subscribes to the write-ahead log to stay current, and promotes the prewarmed node to primary—all without additional cost or user configuration. Read/write endpoints in all regions use this process by default; read-only endpoints are coming soon.