Qortex
Collective memory & governance for fleets of domain-specific agents
Qortex
Collective Memory
A memory framework which aims to define the bedrock of information in which coordinating agents rely on.
Overview
Qortex is an experimental memory framework which attempts to mimic how collectives use, organize and learn as they operate towards their goals.
It uses a custom derived "Memory Band" architecture built for an analogy of my understanding of how collectives function.
The core idea - four-band memory
Governance of agent memory is just as important as retrieval.
- EmbodiedWhat an agent is — skills, experience, intuition, heuristics, capabilities, guardrails, etc.
- PersonalWho the agent is coordinating with — memory scoped to an individual agent/human, so context about an entity carries across sessions.
- ExternalizedWhat the agent has recorded — documents and archived artifacts with vector-indexed recall, shared and retrievable across the collective.
- InstitutionalizedWhat the organization holds — codified policies and SOPs. Update requires a Collective Alignment Session: a proposal, an input window, and a human-approved commit.
Adjacent projects
These are related projects which uses a subset of the same memory system as Qortex, early, personal and indev.
Qortex Visio
An obsidian plugin for allowing LLMs to participate in mind mapping session directly in the canvas through a custom Visuospatial Engine.
Qortex Impromptu
An extension of Visio which uses it's Visuospatial Engine to perform refinement rounds on a handoff prompt
Operating a fleet of agents that need to learn without forgetting their rules?
I'm taking on a small number of design partners — particularly in regulated verticals where governed rule evolution is a hard requirement, not a curiosity.