The breakdown of the heavy hitters #
I've spent a lot of time comparing these based on deployment flexibility and TypeScript integration. Here is how they actually stack up in a real-world AI workflow or web project:
Payload CMS: Best for TS purists. It uses a code-first approach where the schema is defined in TypeScript. The v3 update is huge because it integrates directly into the Next.js App Router, meaning your CMS and frontend can live in one repo.Strapi: The "industry standard" for those who want a GUI. It's great for non-technical users to manage the schema, but the TypeScript depth is shallower compared to Payload or Keystone.Directus: Essentially a database wrapper. It doesn't "own" your data in a proprietary way; it just provides a powerful admin UI on top of your SQL database. It's the best choice if you already have a database and just need an API/Admin panel.Keystone: Very similar to Payload in its "code-first" philosophy and deep TypeScript integration, though it lacks the same level of cloud-managed polish.Sanity: The only one here that isn't truly self-hosted. You get a TypeScript-based "Studio," but the data lives in their content lake. The trade-off is a massive gain in image handling (CDN + hotspots) and a generous free tier.
Technical Trade-offs #
If you're looking for a practical tutorial on which one to pick, look at your database requirements. Directus and Strapi are great for standard SQL setups. However, if you want a seamless AI workflow where your types are automatically generated from your schema to prevent runtime errors in your LLM agent or frontend, Payload and Keystone win.
Payload's approach to the database is particularly interesting. You can run it on a cheap VPS for a few bucks a month, but you're responsible for your own Postgres backups. If you hate DevOps, Sanity is the path of least resistance.
Here is a look at how a basic collection is defined in Payload to show that "code-first" feel:
// payload.config.ts — minimal collection with typed fields
import { buildConfig } from 'payload/config'
import { postgresAdapter } from '@payloadcms/db-postgres'
export default buildConfig({
db: postgresAdapter({
pool: { connectionString: process.env.DATABASE_URL },
}),
collections: [
{
slug: 'posts',
fields: [
{ name: 'title', type: 'text', required: true },
{ name: 'body', type: 'richText' },
],
},
],
})
Final Verdict #
For most new projects, I'm leaning toward Payload v3 because the Next.js integration removes the need for a separate backend server. If you need a purely managed experience with zero infra overhead, go Sanity. If you have a legacy SQL database that needs an API overnight, Directus is the answer.
Next Stop letting your LLM make the final call on your data →
a practical ChatGPT prompt guide, with plenty of directly applicable cases.