We recently published a package on NPM called @neondatabase/toolkit
(@neon/toolkit
on JSR). The idea behind it was to merge the already existing @neondatabase/api-client
and @neondatabase/serverless
packages into a single SDK that is easier for AI agents to consume.
AI agents do better when their search space is less wide.
For example:
- Vercel’s v0 AI product for UI generation works very well because it is limited to React, Tailwind CSS and ShadCN/UI (as the “component library”).
- Replit Agents only currently support Postgres via Neon which is also important for their efficacy in generating good apps.
If these agents had to support a bunch of different frontend frameworks, or various storage providers, they wouldn’t work as well. This might become less relevant in the future as the models get better, but just like specialized software engineers can generally outperform generic engineers when operating with their comfort stack, the same might be true for all AI agents going forward.
Neon already has unique features perfect for AI agents, but we realized that a simple JavaScript script to launch and connect to a Postgres instance required two dependencies:
So, with @neondatabase/toolkit
, we’ve narrowed down the API to make it easier for AI agents to consume it:
Since you can access the underlying toolkit.apiClient
, you can already perform any action from the Neon API. However, we’ll be streamlining more features through the higher-level API such as branching and autoscaling configuration. Furthermore, we’ll soon expose an easier way for consumers to dictate whether they wish to connect via HTTP, WebSockets or good old TCP. (The default will continue to be HTTP.)
Building Agents? Try it. Join us.
Call for Design Partners
Please file an issue in the repository, or reach out to us via Discord to give us any feedback. We’re especially interested in enabling AI agents with this package, but we know that other use cases will emerge as well, and that’s fine!