@detiq/app-brain — The Moat

The Zeta Core is the project's permanent, structured, exportable knowledge.

Grounding (the important bit)

grounding.ts defines how the AI gets real facts before generating tests.

• Default: DirectLoadGrounding — fetch the exact facts for a screen and load them directly. Simple, precise, cheap. Best for a closed workspace.
• Later: VectorSearchGrounding (RAG) — added ONLY for oversized projects whose knowledge is too big to load directly. Same Zeta Core, smarter finder. Agents don't change — only the strategy swaps.

Rule: agents call ground(...) and never invent elements. If a fact isn't in the returned context, the agent asks instead of guessing.