Develop a new version of the engine

Create from scratch a minimal engine that is easy to enhance using AI. The core value is the ability to quickly add the required feature using a ready-made foundation and open examples.

We want to achieve this through the following principles:

1. End-to-end typing: database → server → GraphQL → frontend.
   Prisma types database operations, Pothos handles GraphQL and custom resolvers, and codegen handles client operations. Types should be inferred and generated wherever possible, without manual duplication. This provides the AI with explicit contracts and compilation errors to verify changes. We are considering Prisma 8 for the new project.

2. Functionality in pluggable modules.
   A module should combine everything necessary for its feature: data models, server logic, API, interface, and agent actions. The goal is to add capabilities without modifying the entire engine and without rewriting the core. The specific plugin system has yet to be designed.

3. Minimum code for modifications.
   General mechanisms are provided by the engine; an extension primarily describes its specific task. There should be no need to manually repeat types, write boilerplate glue layers, or maintain multiple descriptions of a single entity. At the same time, code reduction should not be achieved through hidden magic that is difficult for the agent to understand.

4. A structure understandable by AI without knowing the full project history.
   Predictable file locations, explicit dependencies, typed extension points, and short instructions should allow the agent to study the required module and make a local change.

5. A lightweight and minimalist frontend.
   Out of the box, a small shell with essential interface elements. Pages and features are added by modules; unused functionality should not bloat the client. Keeping Next.js is not mandatory: we are considering dropping it in favor of a custom modular frontend if it reduces complexity and dependencies.

6. Open cases as working examples of extensions.
   Each case should contain a task, specific changes, and a way to verify the result. The AI should be able to take such an example, adapt it, and quickly run an experiment. Running MODX or an arbitrary Docker project are examples of extensibility, not mandatory core features.

Result Criterion

A third-party AI, using the documentation and an open case, adds a new feature from the database to the interface, verifies it, and does not require constant prompting from the engine's author.

Distribution Goal

Popularity and engagement: so that people try the engine, enhance it, and share their modules and cases. Sales and monetization are not being designed yet.