PROVIDER Β· Serverless Postgres
Neon
Neon is serverless Postgres with separated compute and storage, offering second-scale database branches, scale-to-zero, and pgvector for development copies and fluctuating traffic.
4.5pts
Pros
Cons
Product Information
Detailed Notes
Overview
Neon is a US serverless Postgres service separating compute and storage, where storage persists pages durably and compute starts and stops on demand, scaling to zero when idle. Its signature feature is database branching, copying data for development and preview environments like Git.
For Backend & Database work it fixes two long-standing Postgres problems, giving development environments real production data through a branch created in seconds, and cutting cost by scaling idle databases to zero. Sites using pgvector for RAG also skip running a separate vector store.
The limits are cold starts and ecosystem binding, since the first connection after scaling to zero adds noticeable latency and services sensitive to time-to-first-byte notice it, while branches and autoscaling bill by compute time and storage beyond quotas. Ownership and product lines have shifted, so confirm regions and pricing in current docs.
Core Strengths
- Separated compute and storage with scale-to-zero and billing by actual compute time
- Branches copy production data in seconds for development, testing, and previews
- Full Postgres with pgvector and common extensions
- Standard Postgres connections keep ORMs and tools working unchanged
Product Ecosystem
Branch workflow
Create a branch per pull request or environment and delete it after use, wiring branch creation into CI so dozens of branches do not sprawl.
Autoscaling and read replicas
Compute scales on demand with read replicas for read-heavy work, sharing storage with the primary for natural consistency.
pgvector and extensions
Vectors live in Postgres for similarity search, fitting small to mid-size RAG while large corpora may still need a dedicated vector store.
Drawbacks
- Scaling to zero adds cold-start latency on the first connection, so latency-sensitive services need a minimum compute
- Branches and autoscaling bill compute time plus storage, and many branches inflate cost
- Cross-region and compliance capabilities should be confirmed against current docs
- Recent changes in ownership and product direction warrant tracking official announcements
Use Cases
- Dev environments needing real data (β β β β β ) - Branches copy production in seconds
- Fluctuating traffic (β β β β ) - Scale-to-zero cuts idle cost
- RAG and vector search (β β β β ) - pgvector removes a separate vector store
- Latency-sensitive services (β β β ) - Cold starts require a minimum compute
Pricing
| Plan | Price | Scope |
|---|---|---|
| Free tier | $0 | Limited compute hours and storage |
| Usage tiers | Compute hours plus storage | Includes branches and autoscaling |
| Enterprise | Custom quote | Dedicated resources and compliance |
Billing combines compute hours and storage with a free tier for validation. Branches consume resources, so set retention policies and keep minimum compute on production to avoid cold starts.
FAQ
-
Neon or Supabase? Neon for branching and scale-to-zero, Supabase for an integrated BaaS with auth, storage, and realtime. Both are Postgres, so migration cost sits in surrounding features.
-
How slow are cold starts? It depends on configuration and driver, though the first query is clearly slower than a warm connection. Keep a minimum compute for API services or warm connections through a pooler.
-
Does it work with common ORMs? Yes over standard Postgres, though serverless functions hit connection limits, so use built-in or external pooling.