Neon review

Neon is a hosted serverless Postgres backend, not a database library or a self-hosted server. It sits in the application data layer: your app uses Postgres, while Neon manages database infrastructure and provides services around it. The platform is part of Databricks.

Its Lakebase Postgres architecture separates compute from object storage. Compute scales with application load and can scale to zero. Neon also offers copy-on-write branches for isolated development and testing environments, managed user authentication based on Better Auth, Functions that run close to the database, S3-compatible Object Storage, and an AI Gateway for frontier and open-source models.

That package can reduce the number of backend services an app needs, but it also puts more of the stack behind one provider. Teams adopting Neon should assess the Postgres connection and their use of provider-specific branches, auth, Functions, storage and AI Gateway separately. The pricing page lists resource-based charges for several of those components.

The integration surface is Postgres, alongside an AI Gateway API, S3-compatible Object Storage and a CLI command shown on the homepage: npx neon init. The homepage also directs developers to Neon’s docs. It does not name programming languages or SDKs, so confirm current SDK support and connection instructions in the vendor’s docs before choosing a client library.

For application data access, plan around Postgres rather than a proprietary query API. The platform’s additional services have their own integration surfaces: Better Auth-based managed authentication, Functions close to the database, S3-compatible storage, and one AI Gateway API for frontier and open-source models. Neon says its Functions run without timeouts, but teams should check execution limits and runtime details in the docs.

Branching is the main development workflow feature: Neon describes isolated, copy-on-write backend environments for testing and development. Confirm how branches fit your migration, seed-data and deployment workflows, and check the docs for current CLI commands and supported integrations.

The Free plan costs $0 and includes 100 projects, 100 CU-hours monthly per project, 1 GB of storage per project and up to 2 CU (8 GB RAM). It is a useful starting point, but compute and storage caps matter once an app has sustained production traffic.

Launch is usage-based, with a listed typical spend of $15 per month for intermittent load and 1 GB. It includes 100 projects, charges $0.106 per CU-hour and $0.35 per GB-month of database storage, and supports up to 16 CU (64 GB RAM). Scale lists a typical $701 monthly spend for high load and 100 GB; it includes 1,000 projects, charges $0.222 per CU-hour and $0.35 per GB-month, and supports up to 56 CU (224 GB RAM). These spends are estimates, not fixed fees.

Paid plans add costs for extra branches, history, backups, Object Storage and Functions. Both include 500 GB network transfer per project, then charge $0.10 per GB. AI Gateway is free during beta, but compute and egress are billed.

Neon’s product overview does not state a software license or a supported self-hosting deployment. Ask Neon to confirm both before treating the service as open source or planning to run it outside Neon’s hosted platform.

Postgres is a familiar database foundation, which can help limit lock-in at the data layer. But the overview does not establish Neon’s export process or the exact steps for moving a production database. Verify how to export database contents and roles, restore them elsewhere, and preserve any required point-in-time history.

Neon-specific branching, managed authentication, Functions, Object Storage and AI Gateway integrations can create additional migration work. The S3-compatible storage interface may ease storage integration, but confirm whether your data and configuration can be exported in the form your next provider needs. Keep application logic portable where possible, and test a migration plan before relying on these services in production.

Neon says compute scales with application load and can scale to zero. Its homepage says the database monitors load ten times a second and that 99.99999% of reads are served from high-performance local caches. Those are vendor claims, not a substitute for an availability commitment: get the applicable service-level agreement, recovery terms and support response times from Neon before relying on it.

Check production capacity against plan ceilings: Launch lists up to 16 CU (64 GB RAM), while Scale lists up to 56 CU (224 GB RAM). Model compute-hours, database storage, branches, history, backups, Functions and network transfer against expected traffic. Transfer above 500 GB per project is billed, and Scale private network transfer costs $0.01 per GB.

The product overview does not establish request limits, security-control details or data-export procedures. Confirm those with Neon, including how authentication, backups and recovery work for your architecture. Test autoscaling and scale-to-zero behavior against your app’s latency and connection requirements.