July 2025 Summaries

App.build is transitioning from a command-line interface to a more accessible web interface that allows users to build and deploy React, Laravel, or FastAPI applications without configuration. This open-source AI agent, initially designed as a reference implementation for agent builders, supports multiple tech stacks including React with tRPC, Laravel (PHP), and FastAPI with NiceGUI. With the discontinuation of the managed app.build version, users can still access the source code for reference and implementation examples. The new web UI aims to enhance user experience by offering features like app previews and supporting multiple deployment platforms, while the CLI will soon be deprecated due to maintenance constraints. The team is focused on expanding tech stack options and deployment platforms, welcoming feedback and contributions via GitHub or Discord.

At Laracon 2025, a new feature for generating Laravel applications through natural language prompts was introduced via app.build, an open-source reference architecture for agent code generation. This tool, initially focused on React/Node templates, now supports Laravel, enabling users to generate full-stack web apps by simply describing their app in a prompt. The process includes scaffolding, writing tests, deploying to the cloud, and integrating a Postgres database, all managed on GitHub. Originating from a research project by Neon, app.build aims to enhance platforms using AI agents and codegen tools. The new Laravel template facilitates rapid prototyping and app creation, aligning with Laravel's best practices, and encourages community contributions under the Apache 2 license.

Neon, a serverless Postgres platform, is being utilized in various AI workflows, offering solutions for vector-based applications, developer tools, and backend support for agents. The platform integrates the open-source pgvector extension, enabling developers to store vector embeddings alongside traditional data for efficient similarity searches and reduced overhead. With features like autoscaling, branching for safe experimentation, and multi-project architecture, Neon enhances performance and isolation for AI applications. Additionally, Neon's MCP Server provides structured API access for database management in developer environments, supporting tools like GitHub Copilot and simplifying tasks such as query tuning and performance monitoring. Neon's serverless capabilities also make it ideal for agentic workflows, allowing fast database provisioning without sign-up requirements, and ensuring cost efficiency through compute suspension and quota controls. This flexibility and ease of integration have made Neon a preferred choice for developers working on AI projects, offering a scalable and economical solution for embedding vector stores, leveraging AI-assisted tools, and powering dynamic agent deployments.

Arseni Kravchenko's blog post offers practical insights into developing production AI agents by emphasizing system design and software engineering over finding perfect prompts or frameworks. It outlines six key principles, including investing in clear and detailed system prompts, effective context management, and the careful design of tools and feedback loops. The post highlights the importance of LLM-driven error analysis and warns against assuming frustrating agent behavior is due to model flaws, as it might stem from system errors such as missing tools or ambiguous prompts. By focusing on reliable and recoverable agents, developers can iteratively improve performance and ensure robust AI systems.

Postgres has evolved beyond a traditional database to become a platform capable of orchestrating entire infrastructures through event-driven systems, thanks to features such as event triggers, LISTEN/NOTIFY, and transactional DDL. These capabilities transform schema changes from mere DDL statements into reliable, observable, and actionable events, allowing for automated responses to structural changes without the need for additional scaffolding like message queues or cron jobs. By utilizing Postgres's event-driven arsenal, including event triggers and the outbox pattern, developers can build systems where schema changes automatically trigger downstream updates such as regenerating SDKs or syncing analytics, exemplified by the logistics SaaS, FleetOps. This architecture, supported by Neon's serverless infrastructure, ensures transactional consistency, scalability, and reduces external dependencies, making Postgres a central command center for automated processes.

Devin, an AI software engineer from Cognition Lab, has expanded its capabilities by integrating with Neon's MCP (Model Context Protocol) server marketplace, which allows direct interaction with Neon databases and other development tools. A practical example demonstrates Devin's ability to address a performance bottleneck in an e-commerce website's API by using its tools to analyze, optimize, and improve query performance. Although the initial optimization attempt with indexing failed, Devin successfully implemented a materialized view that enhanced performance by 97%, while also addressing data staleness concerns by recommending regular view refreshes using the pg_cron extension. This integration enables Devin to take on complex tasks and safely test optimizations on new branches, safeguarding data integrity. The expansion of MCP server integrations broadens Devin's capabilities, allowing interaction with tools like Sentry, DataDog, Linear, and Slack, thereby increasing the range of challenges it can tackle.

Neon Local Connect is a newly launched VS Code extension designed to simplify the process of connecting Neon database branches to local development environments by utilizing a static localhost connection string, eliminating the need for separate connection strings or CLI setups. This integration allows developers to manage databases directly from their IDE, offering functionalities like browsing tables, creating or resetting branches, launching ephemeral test environments, and running queries. The extension works by using a Docker-based local proxy to route traffic to Neon database branches, creating an experience akin to using a local Postgres instance but with the advantages of Neon’s serverless cloud and branching engine. Developers can quickly connect to or create new branches, ensuring isolated and clean environments for testing without risking production data. This tool aims to enhance local workflows by offering easy branch management, automated environment resets, and the ability to test migrations in isolation, all within the familiar VS Code interface.

app.build, originally focused on creating apps with a fixed tech stack, has evolved to support data applications using a modern Python stack with the NiceGUI framework, which excels in data visualization and offers flexibility through Python libraries. The transition allows for the development of internal enterprise apps like dashboards and tech demos, using PostgreSQL for persistent storage and ensuring production-readiness similar to FastAPI. The new stack, offering better functionality than previous tRPC apps, can leverage integrations like Databricks Unity Catalog, although these remain optional to accommodate customizations. Despite the discontinuation of its managed version, app.build's source code is available for developers interested in building agents or exploring implementation examples. The platform aims to continue expanding its offerings by introducing additional stacks, with a focus on maintaining quality through structured scaffolding and validation processes.

Vectstore, an AI search platform built on Rust, faced challenges with their initial setup using AWS RDS for relational data and Pinecone for vector search due to complexity, high latency, and rising costs. To address these issues, they switched to Neon, a serverless Postgres database with the pgvector extension, which allowed them to consolidate both relational and vector data into a single system. This transition resulted in reduced latency from 200ms to 80ms, simpler architecture, and a smoother developer experience across multiple regions. Neon's features such as branching, autoscaling, and native vector search support streamlined their operations, decreased infrastructure costs, and improved overall performance by enabling efficient management of multi-region deployments and spiky workloads.

In May and June 2025, the Neon platform faced stability challenges due to an unexpected surge in database and branch creation, primarily driven by agentic AI partners, which far exceeded initial projections. This resulted in a series of incidents that affected database operations, with a significant increase in operational load leading to service degradation. The company identified scalability limits with its Kubernetes-based infrastructure and a control plane database under strain from increased metadata handling. To address these issues, Neon implemented a horizontally-scalable architecture known as "Cells" and enforced stricter workload limits. They also plan to isolate critical components to improve system resilience and predictability. Despite the incidents, Neon has taken steps to enhance its infrastructure, aiming to achieve better reliability and customer satisfaction in the future.

Neon's recent enhancements to its storage performance have focused on reducing latency, improving throughput, and optimizing resource usage. By implementing smarter sharding, the company has achieved a significant reduction in WAL (Write-Ahead Log) decoding work, resulting in a 2x increase in ingest throughput for large tenants. Additionally, compressing WAL transmission has decreased network bandwidth usage by 70%, while faster disk writes on Safekeepers have led to a 230% increase in WAL write throughput. Other improvements include reduced latency for sequential and index scans through page prefetching, which sped up schema finalization by 300%, and more responsive L0 compaction to lower read amplification and latency. The introduction of parallel, pipelined S3 uploads has doubled upload throughput, cutting ingestion benchmarks by 30 minutes. Finally, an improved shard split policy offers an 18% speedup in ingestion for medium-sized tenants and better load balancing for larger ones, laying the groundwork for continued performance enhancements in Neon's storage system.

pg_cron is a Postgres extension that simplifies the scheduling of database tasks using familiar cron syntax, allowing for automated data transformations, notifications, and maintenance within the database environment. However, issues can arise when scheduled tasks fail due to schema changes, such as tables being dropped without corresponding updates to the cron jobs, leading to persistent error logs. The architecture of pg_cron, which operates by storing job definitions as raw SQL text, means it does not automatically detect changes in the database schema, resulting in recurring failures when a table is removed. To address this, Postgres event triggers can be employed to automatically unschedule affected pg_cron jobs upon detecting a table drop, effectively preventing error spam without manual intervention. This solution leverages the sql_drop event trigger to maintain synchronization between pg_cron's scheduling metadata and the actual state of the database schema, offering a self-healing mechanism that enhances operational reliability. While pg_cron is ideal for tasks like database maintenance and lightweight ETL within the same database, other tools like Apache Airflow or Inngest may be more suitable for tasks requiring cross-database coordination, long-running jobs, or complex dependency graphs.

Neon, a modern OLTP database system, separates storage and compute to offer elasticity and serverless operation, allowing users to pay only for what they use without committing to specific storage or compute capacities. This architecture typically incurs latency due to additional network hops, but Neon mitigates this with a Local File Cache (LFC) that uses NVMe storage to provide performance comparable to local disks. The LFC, acting as an intermediary cache layer, is flexible and resizable, allowing users to optimize for workload needs without sacrificing elasticity. Benchmark tests, including a TPCC-like benchmark, demonstrate that Neon maintains high query performance and low latency without requiring overprovisioned local storage, showcasing its ability to deliver both elasticity and robust performance. Recent developments have made large-cache configurations publicly available as part of the Databricks Lakebase launch, and future updates aim to enhance the LFC's capabilities further.

Modern software development has evolved rapidly, but relational databases have lagged behind, traditionally maintaining a cumbersome and risky reputation. This has become increasingly problematic with the rise of AI agents that demand faster, more agile infrastructure than conventional databases can offer. In response, Neon has developed a revolutionary approach to relational databases by introducing branching as a core primitive, modeled after version control systems used in software development. This method allows developers to create instant, cost-efficient, and safe environments for experimentation, without the fear of disrupting the primary database. Unlike traditional systems that rely on duplicative and slow processes, Neon's architecture supports efficient, copy-on-write branching, enabling rapid iteration, easy rollbacks, and seamless integration into modern workflows, thus transforming the interaction between developers, agents, and databases. This innovation not only enhances productivity by reducing the barriers to testing and development but also positions branching as a potential standard for future database management, where iteration is swift and errors are easily reversible.

The text introduces a template designed to simplify the creation of internal tools using Neon, StackAuth, and Vercel. It highlights the common challenges faced by tech companies, particularly smaller organizations, in managing internal tools, often resulting in insecure or overly complex solutions. The template offers a streamlined approach, providing a preconfigured setup with a Postgres database, API routes, authentication, and an admin panel, allowing users to rapidly deploy internal applications. It emphasizes the ease of setup and customization, facilitating access control through StackAuth's role-based permissions and domain whitelisting, while Neon provides a serverless Postgres database ideal for integration with Next.js. Vercel aids in hosting the app, making the deployment process straightforward. The template aims to offer a secure, scalable, and cost-effective solution, enabling developers to focus on their core work without the need to manage a full-scale production environment.

The blog post explores building a full-text search Content Management System (CMS) using Neon, a serverless Postgres platform, and the pg_search extension. It highlights the potential of using databases directly for CMS purposes by bypassing traditional CMS platforms, thereby simplifying architecture and improving performance. The guide walks through setting up a Neon account, creating projects, and integrating Prisma for database management. It explains the implementation of full-text search capabilities using pg_search and BM25, a search algorithm that ranks results by relevance. The system is enhanced with features like typo tolerance, phrase search, and adaptive fuzzy matching, ensuring a robust search experience. The post concludes by illustrating the versatility and simplicity of combining Neon and pg_search, suggesting possibilities for various applications beyond a basic CMS, such as e-commerce platforms and knowledge bases, without the need for complex infrastructure or dedicated search engines.

Neon has introduced several new features and improvements to its MCP Server, enhancing its functionality and ease of use for AI agents and developers. The server, which provides a streamlined interface for tasks like database management and query inspection, now includes a reference catalog for supported actions, a one-click integration with the Cursor platform, and direct connection capabilities without the need for the mcp-remote bridge. Additionally, the server supports API key authentication and streamable HTTP for environments where Server-Sent Events (SSE) are less effective. New actions have been added to enhance performance and visibility, such as inspecting compute resources and diagnosing slow queries, while quality-of-life improvements include smarter project listing, more informative authentication dialogs, and detailed organization metadata in project actions. These updates aim to make the Neon MCP Server more efficient for managing multi-organizational setups and improving workflow connectivity.

Neon now supports Postgres Event Triggers through the neon_superuser role, providing users with the ability to monitor schema changes, automate workflows, and enhance sync engines like Zero. Event triggers in Neon function similarly to those in standard Postgres, allowing them to respond to Data Definition Language (DDL) commands such as CREATE, ALTER, and DROP. This feature enables real-time tracking of schema alterations, facilitates automation of maintenance tasks when schema changes occur, and supports the synchronization of local and remote databases by platforms like Zero. Additionally, event triggers can enforce production guardrails by restricting certain DDL operations, and they can manage pg_cron jobs automatically to prevent errors from orphaned tasks. This new layer of control is aimed at users who have been limited by the absence of event triggers in other managed Postgres platforms, with support available by default for anyone using the neon_superuser role.

Neon has partnered with TanStack as their official database partner to enhance the developer experience within the TanStack ecosystem, particularly with the anticipated launch of TanStack Start 1.0. Vite's popularity as a foundational technology has led Neon to create an official add-on for Create TanStack, which allows developers to swiftly set up fullstack applications using a Neon Postgres instance. The introduction of Neon Launchpad streamlines the integration of Postgres into JavaScript projects by enabling instant access to a Postgres instance without requiring a sign-up, a feature that has already been adopted by platforms like Sentry and Netlify DB. Neon emphasizes its commitment to open source, providing packages like NeonDB and a Vite plugin to facilitate seamless Postgres integration, further empowering developers and supporting the Vite and TanStack communities.

Traconiq, a company specializing in managing logistics for complex vehicle fleets, faced increasing costs and operational challenges with Amazon RDS as their telemetry dataset grew to multiple terabytes. The costs of maintaining separate RDS instances across multiple regions and environments, as well as the static nature of RDS storage and compute, became prohibitive. Traconiq transitioned to Neon, a fully managed Postgres platform that separates storage and compute, allowing for branching and autoscaling to match actual workload demands. This change enabled Traconiq to eliminate the need for multiple RDS instances, optimize costs by scaling compute according to daytime spikes and nighttime lulls, and improve recovery times with Neon's branching capabilities. Additionally, they migrated a geo lookup service, benefiting from concurrent branch operations that would have been cost-prohibitive with RDS.