neon inspect db ships with 14 subcommands covering table bloat, unused indexes, slow queries…
Every result is capped and structured (JSON/YAML). No need to give raw SQL access or connection strings to your agent.
Learn how to use it:
What do you do when someone (or an agent) drops a table in production?
We put together a full guide on point-in-time restores, snapshots, and selective table recoveries:
We recently made Lakebase Search available on Neon: hybrid vector and full-text search built for agents.
One half of that is lakebase_text. Here's why we built it, and why Postgres' default BM25 solution isn't enough 🧵
The index stores durably on Neon's object storage layer, the same as the rest of your data. This makes scale-to-zero works, together with branching:
When you branch a Neon database, you get an instant copy of the lakebase_bm25 index alongside the data, so you can test new
lakebase_text ships as part of Lakebase Search, alongside lakebase_vector for ANN vector search
Both live in the same Postgres: you get hybrid search (BM25 + vector similarity, joined against your operational tables, filtered by tenant) in a single SQL query.
A year ago we split @neondatabase infra into a cell-based architecture to scale further per region. At first we were adding 1 cell per month. We're now adding 8 per month. Incredible work by our infra team.
What shipped this week:
- Support for passkey login
- A neon-diff command that shows your agents changes between branches
- Expanded private networking capacity in three AWS regions
- Object Storage, Functions, AI Gateway are beta
All in our Friday changelog:
We started by building a serverless Postgres database, with separate compute and storage and branches that unlock development speed. Now that developers are using agents as the main interface to build, we're building a backend with the same principles 🧵
AI Gateway gives you one credential and one bill for a wide catalog of frontier and open-source models, powered by Databricks
The experience it unlocks: your models under one bill (Neon's) without markup on model pricing
Building a backend like this takes more work vs acquiring a bunch of services and bundling them together. We need your feedback to take these tools to GA - tell your agent to deploy a full Neon backend, and tell us what breaks: