Learn how Neon's autoscaling works - it estimates Postgres' working set size and keeps it in memory. Engineering post here

Changelog

The latest product updates from Neon

Select the right compute size with predefined compute configurations

We've updated the Edit compute and Add new compute drawers in the Neon Console to include predefined compute configurations. With recommended min and max autoscaling settings and autosuspend timeout periods, these configurations make it easier to select the right compute size for your needs.

Compute configurations

Computes with more vCPU and memory

We've added support for larger computes with more vCPU and memory. If you're a Neon Scale plan user, you can now scale your computes up to 9 vCPU with 36 GB of RAM or 10 vCPU with 40 GB of RAM.

Larger compute sizes

More branch protection

If you've been following along, you know that we recently added safeguards to prevent you from deleting protected branches. This week, we're extended that protection:

  • You can no longer delete a Neon project that has protected branches.
  • You can no longer delete computes that belong to protected branches.

The protected branches feature is available with the Neon Scale plan.

Hackable AI starter apps

AI app banner

We've published a new set of hackable, pre-built AI starter apps to help you get up and running with Postgres as your vector store. The set includes different types of AI applications, including:

  • AI chatbot
  • RAG chatbot
  • Semantic search
  • Hybrid search
  • Reverse image search
  • Chat with PDF

Clone an app and make it your own.

For more, check out our new Neon AI Starter Kit, where you'll find links to our starter apps, docs, and a collection of AI applications built with Neon.

Consider sharing your AI app on our #showcase channel on Discord. We'd love to see what you're building.

Fixes & improvements
  • In a Postgres primary-standby configuration, certain settings should be no smaller on a standby than on the primary in order to ensure that the standby does not run out of shared memory during recovery, as described in the PostgreSQL hot standby documentation. For Neon read replicas, it's no different. The same settings should be no smaller on a read replica compute (the "standby") than on the default read-write compute (the "primary"). For this reason, the following settings on read replica computes are now synchronized with the settings on the default read-write compute when the read replica compute is started:
    • max_connections
    • max_prepared_transactions
    • max_locks_per_transaction
    • max_wal_senders
    • max_worker_processes
  • Fixed the SQLAlchemy code snippet in the Connection Details widget in the Neon Console. The host variable was missing a comma.
  • We've made it even clearer in our error message that lets you know when you've exceeded the permitted number of concurrently active computes. Neon has a default limit of 20 concurrently active computes to protect your account from unintended usage. See connection errors for more information.
  • Fixed an issue with the List projects API that caused it to return an empty result set when including an org_id value.
  • Fixed an issue that caused the Create project page to be displayed if a Project page could not be shown due to an error. The Projects list page is now shown instead.
  • For Free Tier users, the Data Transfer metric in the Resources remaining widget on the Project Dashboard now shows 0 when the 5 GB allowance is exceeded, indicating that the allowance has been fully used.
  • Console navigation was improved by carrying forward the branch and database selected on the Dashboard when navigating to other console pages.
  • Console themes (System, Light, Dark) are now set through the Profile menu in the Neon Console. The Theme page, previously accessible from the Settings page, has been removed.
  • Fixed an issue with the rum extension that caused an error when building a RUM index with a large amount of data.
  • Fixed an issue with project sharing where an existing Neon account could not access a shared project after changing their email address to the one the project was shared with.
Back to all changelog posts