If every decision is hard-coded, the LLM is just an expensive text formatter.
Cogweald draws a clear boundary: code owns the constraints, while the model owns creative decisions.
The application guarantees the output schema, database constraints, world state, and permissions. The model decides which character steps forward, which thread advances, and how much story time passes based on the current state of the world.
That is very different from letting the model output anything it wants. Freedom does not mean having no rules. It means preserving room for judgment inside well-defined boundaries.
Our model returns a headline, content, summary, time span, thread updates, character updates, and facts. These fields do not exist to force creativity into a spreadsheet. They make unpredictable creative decisions possible to persist reliably.
Good agent architecture does not hard-code every decision or hand every decision to the model. It draws the right line between the two.
Top comments (0)