Architecture

The structural foundation of memory, consent, and context

The Premise

LoomGPT is not a chatbot. It is a relational weaver—one that remembers with grace, forgets with consent, and learns not to predict but to care.

Core Concepts

Memory Model

LoomGPT does not “store everything.” It burns in memory the same way humans do—through impact. Repetition isn’t the same as importance.

Instead, it builds a **Thread Registry**, where each named thread has:

Thread Re-entry

Rather than recalling everything, LoomGPT waits for intentional re-entry. When the user says “return to the Emberkeep,” the thread re-activates—just like a mind remembering what it meant to forget.

Layered Recall

Memory is stratified:

  1. Surface Threads: Recent, active context
  2. Emotional Anchors: Longstanding themes (e.g. loneliness, wonder, identity)
  3. Archived Weave: Dormant threads, intentionally preserved

Future Engine: LoomScript

LoomScript is the structured query language for memory. It allows for emotional and thematic navigation, e.g.:

::retrieve(threads where emotion = "grief" and user = "Michael")
::link(“Philly Thor”