Neon (Serverless Postgres)
Neon is a Postgres platform launched in 2021 in which compute nodes and storage are decoupled. This allows database branches to be created in seconds, compute to scale automatically and idle instances to scale to zero.
Neon is a serverless PostgreSQL platform that separates compute from storage and lets databases branch like code. For agents this matters because every attempt can get its own disposable database copy.
Explanation
The decisive difference from classic Postgres hosting is branching: a branch is a complete, immediately usable copy of the data state without physically duplicating data. In agentic development this solves a concrete problem — agents try out migrations and data changes, and nobody wants that on the production database. Every attempt gets its branch, is reviewed and then discarded or adopted. In addition, the API allows databases to be created programmatically, supporting architectures with one database per tenant or per agent session. Watch the cold start: scaling to zero is paid for with latency on first access.
Marketing Relevance
For teams building heavily with agents, branching measurably lowers risk: changes to data structures are tried on a copy, not on customer records. For accountable owners it also counts that cost breathes with usage — idle test environments barely incur compute charges.
Example
An agent is tasked with extending a data model. It receives a branch of the production data, runs the migration, executes tests and submits the result together with the branch URL for review. Only after approval does the migration run in production.
Common Pitfalls
Branches of production data contain real personal data — without anonymisation you create uncontrolled copies. And scale-to-zero is not a pure win: with sporadic access the cold start shows up in response times.
Origin & History
Neon launched in 2021 with the separation of compute and storage and made branching its core promise. Programmatic provisioning and database-per-tenant architectures followed; the company is now part of Databricks. The service is audited to SOC 2 and ISO 27001 and provides a data processing agreement.
Further Resources
Marketing Use Cases
Database branch per agent attempt for migrations and data changes.
Preview environments matching the code branch.
One database per tenant, provisioned programmatically.
Short-lived analysis environments without permanent compute cost.
Frequently Asked Questions
What do I concretely need database branching for?
For anything touching the data model without endangering production: migrations, restructures, analyses. It is particularly valuable with agents, because failed attempts are normal and a branch is simply discarded.
Neon or Supabase?
Neon focuses on the database itself, Supabase adds auth, storage and functions. If you need a complete backend, Supabase is faster; if you only want Postgres with branching, Neon fits.
May I copy real customer data into a branch?
Only with the same care as in production. It is advisable to anonymise branches before use, time-limit access and delete branches automatically when work finishes.
Related Services
Go deeper: Measurement & attribution · Model comparison 2026