Mnemex

Overview Inspect Schema Related Servers Score Discussions

mnemex
site
natural-language-activation-plan

index.html

index.html•184 kB

<!doctype html> <html lang="en" class="no-js"> <head> <meta charset="utf-8"> <meta name="viewport" content="width=device-width,initial-scale=1"> <meta name="description" content="Memory persistence for AI assistants with temporal decay"> <meta name="author" content="prefrontal-systems"> <link rel="canonical" href="https://cortexgraph.dev/natural-language-activation-plan/"> <link rel="icon" href="../assets/images/favicon.png"> <meta name="generator" content="mkdocs-1.6.1, mkdocs-material-9.7.0"> <title>Natural Language Activation Implementation Plan - CortexGraph Documentation</title> <link rel="stylesheet" href="../assets/stylesheets/main.618322db.min.css"> <link rel="stylesheet" href="../assets/stylesheets/palette.ab4e12ef.min.css"> <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin> <link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Roboto:300,300i,400,400i,700,700i%7CRoboto+Mono:400,400i,700,700i&display=fallback"> <style>:root{--md-text-font:"Roboto";--md-code-font:"Roboto Mono"}</style> <script>__md_scope=new URL("..",location),__md_hash=e=>[...e].reduce(((e,_)=>(e<<5)-e+_.charCodeAt(0)),0),__md_get=(e,_=localStorage,t=__md_scope)=>JSON.parse(_.getItem(t.pathname+"."+e)),__md_set=(e,_,t=localStorage,a=__md_scope)=>{try{t.setItem(a.pathname+"."+e,JSON.stringify(_))}catch(e){}}</script> </head> <body dir="ltr" data-md-color-scheme="default" data-md-color-primary="indigo" data-md-color-accent="indigo"> <input class="md-toggle" data-md-toggle="drawer" type="checkbox" id="__drawer" autocomplete="off"> <input class="md-toggle" data-md-toggle="search" type="checkbox" id="__search" autocomplete="off"> <label class="md-overlay" for="__drawer"></label> <div data-md-component="skip"> <a href="#natural-language-activation-implementation-plan" class="md-skip"> Skip to content </a> </div> <div data-md-component="announce"> </div> <header class="md-header" data-md-component="header"> <nav class="md-header__inner md-grid" aria-label="Header"> <a href=".." title="CortexGraph Documentation" class="md-header__button md-logo" aria-label="CortexGraph Documentation" data-md-component="logo"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a3 3 0 0 0 3-3 3 3 0 0 0-3-3 3 3 0 0 0-3 3 3 3 0 0 0 3 3m0 3.54C9.64 9.35 6.5 8 3 8v11c3.5 0 6.64 1.35 9 3.54 2.36-2.19 5.5-3.54 9-3.54V8c-3.5 0-6.64 1.35-9 3.54"/></svg> </a> <label class="md-header__button md-icon" for="__drawer"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M3 6h18v2H3zm0 5h18v2H3zm0 5h18v2H3z"/></svg> </label> <div class="md-header__title" data-md-component="header-title"> <div class="md-header__ellipsis"> <div class="md-header__topic"> CortexGraph Documentation </div> <div class="md-header__topic" data-md-component="header-topic"> Natural Language Activation Implementation Plan </div> </div> </div> <form class="md-header__option" data-md-component="palette"> <input class="md-option" data-md-color-media="(prefers-color-scheme: light)" data-md-color-scheme="default" data-md-color-primary="indigo" data-md-color-accent="indigo" aria-label="Switch to dark mode" type="radio" name="__palette" id="__palette_0"> <label class="md-header__button md-icon" title="Switch to dark mode" for="__palette_1" hidden> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a4 4 0 0 0-4 4 4 4 0 0 0 4 4 4 4 0 0 0 4-4 4 4 0 0 0-4-4m0 10a6 6 0 0 1-6-6 6 6 0 0 1 6-6 6 6 0 0 1 6 6 6 6 0 0 1-6 6m8-9.31V4h-4.69L12 .69 8.69 4H4v4.69L.69 12 4 15.31V20h4.69L12 23.31 15.31 20H20v-4.69L23.31 12z"/></svg> </label> <input class="md-option" data-md-color-media="(prefers-color-scheme: dark)" data-md-color-scheme="slate" data-md-color-primary="indigo" data-md-color-accent="indigo" aria-label="Switch to light mode" type="radio" name="__palette" id="__palette_1"> <label class="md-header__button md-icon" title="Switch to light mode" for="__palette_0" hidden> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 18c-.89 0-1.74-.2-2.5-.55C11.56 16.5 13 14.42 13 12s-1.44-4.5-3.5-5.45C10.26 6.2 11.11 6 12 6a6 6 0 0 1 6 6 6 6 0 0 1-6 6m8-9.31V4h-4.69L12 .69 8.69 4H4v4.69L.69 12 4 15.31V20h4.69L12 23.31 15.31 20H20v-4.69L23.31 12z"/></svg> </label> </form> <script>var palette=__md_get("__palette");if(palette&&palette.color){if("(prefers-color-scheme)"===palette.color.media){var media=matchMedia("(prefers-color-scheme: light)"),input=document.querySelector(media.matches?"[data-md-color-media='(prefers-color-scheme: light)']":"[data-md-color-media='(prefers-color-scheme: dark)']");palette.color.media=input.getAttribute("data-md-color-media"),palette.color.scheme=input.getAttribute("data-md-color-scheme"),palette.color.primary=input.getAttribute("data-md-color-primary"),palette.color.accent=input.getAttribute("data-md-color-accent")}for(var[key,value]of Object.entries(palette.color))document.body.setAttribute("data-md-color-"+key,value)}</script> <label class="md-header__button md-icon" for="__search"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg> </label> <div class="md-search" data-md-component="search" role="dialog"> <label class="md-search__overlay" for="__search"></label> <div class="md-search__inner" role="search"> <form class="md-search__form" name="search"> <input type="text" class="md-search__input" name="query" aria-label="Search" placeholder="Search" autocapitalize="off" autocorrect="off" autocomplete="off" spellcheck="false" data-md-component="search-query" required> <label class="md-search__icon md-icon" for="__search"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M20 11v2H8l5.5 5.5-1.42 1.42L4.16 12l7.92-7.92L13.5 5.5 8 11z"/></svg> </label> <nav class="md-search__options" aria-label="Search"> <button type="reset" class="md-search__icon md-icon" title="Clear" aria-label="Clear" tabindex="-1"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M19 6.41 17.59 5 12 10.59 6.41 5 5 6.41 10.59 12 5 17.59 6.41 19 12 13.41 17.59 19 19 17.59 13.41 12z"/></svg> </button> </nav> <div class="md-search__suggest" data-md-component="search-suggest"></div> </form> <div class="md-search__output"> <div class="md-search__scrollwrap" tabindex="0" data-md-scrollfix> <div class="md-search-result" data-md-component="search-result"> <div class="md-search-result__meta"> Initializing search </div> <ol class="md-search-result__list" role="presentation"></ol> </div> </div> </div> </div> </div> <div class="md-header__source"> <a href="https://github.com/prefrontal-systems/cortexgraph" title="Go to repository" class="md-source" data-md-component="source"> <div class="md-source__icon md-icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg> </div> <div class="md-source__repository"> prefrontal-systems/cortexgraph </div> </a> </div> </nav> </header> <div class="md-container" data-md-component="container"> <nav class="md-tabs" aria-label="Tabs" data-md-component="tabs"> <div class="md-grid"> <ul class="md-tabs__list"> <li class="md-tabs__item"> <a href=".." class="md-tabs__link"> Home </a> </li> <li class="md-tabs__item"> <a href="../installation/" class="md-tabs__link"> Getting Started </a> </li> <li class="md-tabs__item"> <a href="../architecture/" class="md-tabs__link"> Documentation </a> </li> <li class="md-tabs__item"> <a href="../deployment/" class="md-tabs__link"> Deployment </a> </li> <li class="md-tabs__item"> <a href="../CONTRIBUTING/" class="md-tabs__link"> Development </a> </li> <li class="md-tabs__item"> <a href="../features/auto-recall-conversation/" class="md-tabs__link"> Features </a> </li> <li class="md-tabs__item"> <a href="../LICENSE/" class="md-tabs__link"> About </a> </li> </ul> </div> </nav> <main class="md-main" data-md-component="main"> <div class="md-main__inner md-grid"> <div class="md-sidebar md-sidebar--primary" data-md-component="sidebar" data-md-type="navigation" > <div class="md-sidebar__scrollwrap"> <div class="md-sidebar__inner"> <nav class="md-nav md-nav--primary md-nav--lifted" aria-label="Navigation" data-md-level="0"> <label class="md-nav__title" for="__drawer"> <a href=".." title="CortexGraph Documentation" class="md-nav__button md-logo" aria-label="CortexGraph Documentation" data-md-component="logo"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a3 3 0 0 0 3-3 3 3 0 0 0-3-3 3 3 0 0 0-3 3 3 3 0 0 0 3 3m0 3.54C9.64 9.35 6.5 8 3 8v11c3.5 0 6.64 1.35 9 3.54 2.36-2.19 5.5-3.54 9-3.54V8c-3.5 0-6.64 1.35-9 3.54"/></svg> </a> CortexGraph Documentation </label> <div class="md-nav__source"> <a href="https://github.com/prefrontal-systems/cortexgraph" title="Go to repository" class="md-source" data-md-component="source"> <div class="md-source__icon md-icon"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg> </div> <div class="md-source__repository"> prefrontal-systems/cortexgraph </div> </a> </div> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href=".." class="md-nav__link"> Home </a> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_2" > <label class="md-nav__link" for="__nav_2" id="__nav_2_label" tabindex="0"> Getting Started </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_2_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_2"> Getting Started </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../installation/" class="md-nav__link"> Installation </a> </li> <li class="md-nav__item"> <a href="../quickstart/" class="md-nav__link"> Quick Start </a> </li> <li class="md-nav__item"> <a href="../configuration/" class="md-nav__link"> Configuration </a> </li> </ul> </nav> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_3" > <label class="md-nav__link" for="__nav_3" id="__nav_3_label" tabindex="0"> Documentation </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_3_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_3"> Documentation </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../architecture/" class="md-nav__link"> Architecture </a> </li> <li class="md-nav__item"> <a href="../api/" class="md-nav__link"> API Reference </a> </li> <li class="md-nav__item"> <a href="../graph_features/" class="md-nav__link"> Knowledge Graph </a> </li> <li class="md-nav__item"> <a href="../scoring_algorithm/" class="md-nav__link"> Scoring Algorithm </a> </li> </ul> </nav> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_4" > <label class="md-nav__link" for="__nav_4" id="__nav_4_label" tabindex="0"> Deployment </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_4_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_4"> Deployment </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../deployment/" class="md-nav__link"> Deployment Guide </a> </li> <li class="md-nav__item"> <a href="../security/" class="md-nav__link"> Security </a> </li> </ul> </nav> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_5" > <label class="md-nav__link" for="__nav_5" id="__nav_5_label" tabindex="0"> Development </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_5_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_5"> Development </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../CONTRIBUTING/" class="md-nav__link"> Contributing </a> </li> <li class="md-nav__item"> <a href="../ROADMAP/" class="md-nav__link"> Roadmap </a> </li> <li class="md-nav__item"> <a href="../future_roadmap/" class="md-nav__link"> Future Plans </a> </li> </ul> </nav> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_6" > <label class="md-nav__link" for="__nav_6" id="__nav_6_label" tabindex="0"> Features </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_6_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_6"> Features </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../features/auto-recall-conversation/" class="md-nav__link"> Auto-Recall </a> </li> </ul> </nav> </li> <li class="md-nav__item md-nav__item--nested"> <input class="md-nav__toggle md-toggle md-toggle--indeterminate" type="checkbox" id="__nav_7" > <label class="md-nav__link" for="__nav_7" id="__nav_7_label" tabindex="0"> About </label> <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_7_label" aria-expanded="false"> <label class="md-nav__title" for="__nav_7"> About </label> <ul class="md-nav__list" data-md-scrollfix> <li class="md-nav__item"> <a href="../LICENSE/" class="md-nav__link"> License </a> </li> <li class="md-nav__item"> <a href="../CHANGELOG/" class="md-nav__link"> Changelog </a> </li> </ul> </nav> </li> </ul> </nav> </div> </div> </div> <div class="md-sidebar md-sidebar--secondary" data-md-component="sidebar" data-md-type="toc" > <div class="md-sidebar__scrollwrap"> <div class="md-sidebar__inner"> <nav class="md-nav md-nav--secondary" aria-label="Table of contents"> <label class="md-nav__title" for="__toc"> Table of contents </label> <ul class="md-nav__list" data-md-component="toc" data-md-scrollfix> <li class="md-nav__item"> <a href="#spreading-activation-for-cortexgraph-memory-recall" class="md-nav__link"> Spreading Activation for CortexGraph Memory Recall </a> </li> <li class="md-nav__item"> <a href="#executive-summary" class="md-nav__link"> Executive Summary </a> </li> <li class="md-nav__item"> <a href="#table-of-contents" class="md-nav__link"> Table of Contents </a> </li> <li class="md-nav__item"> <a href="#problem-statement" class="md-nav__link"> Problem Statement </a> <nav class="md-nav" aria-label="Problem Statement"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#current-state" class="md-nav__link"> Current State </a> </li> <li class="md-nav__item"> <a href="#the-gap-natural-conversational-activation" class="md-nav__link"> The Gap: Natural Conversational Activation </a> </li> <li class="md-nav__item"> <a href="#why-this-matters" class="md-nav__link"> Why This Matters </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#research-findings" class="md-nav__link"> Research Findings </a> <nav class="md-nav" aria-label="Research Findings"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#state-of-the-art-2024-2025" class="md-nav__link"> State-of-the-Art (2024-2025) </a> <nav class="md-nav" aria-label="State-of-the-Art (2024-2025)"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#1-imdmr-multi-dimensional-memory-retrieval-nov-2025" class="md-nav__link"> 1. IMDMR: Multi-Dimensional Memory Retrieval (Nov 2025) </a> </li> <li class="md-nav__item"> <a href="#2-spreadpy-spreading-activation-library-july-2025" class="md-nav__link"> 2. SpreadPy: Spreading Activation Library (July 2025) </a> </li> <li class="md-nav__item"> <a href="#3-mem0-scalable-production-memory-apr-2025" class="md-nav__link"> 3. Mem0: Scalable Production Memory (Apr 2025) </a> </li> <li class="md-nav__item"> <a href="#4-a-mem-agentic-memory-with-zettelkasten-feb-2025" class="md-nav__link"> 4. A-MEM: Agentic Memory with Zettelkasten (Feb 2025) </a> </li> <li class="md-nav__item"> <a href="#5-context-window-paradox-industry-research-2025" class="md-nav__link"> 5. Context Window Paradox (Industry Research 2025) </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#synthesis-what-cortexgraph-needs" class="md-nav__link"> Synthesis: What CortexGraph Needs </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#solution-architecture" class="md-nav__link"> Solution Architecture </a> <nav class="md-nav" aria-label="Solution Architecture"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#three-layer-activation-system" class="md-nav__link"> Three-Layer Activation System </a> </li> <li class="md-nav__item"> <a href="#core-components" class="md-nav__link"> Core Components </a> <nav class="md-nav" aria-label="Core Components"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#component-1-activation-trigger-detector" class="md-nav__link"> Component 1: Activation Trigger Detector </a> </li> <li class="md-nav__item"> <a href="#component-2-spreading-activation-engine" class="md-nav__link"> Component 2: Spreading Activation Engine </a> </li> <li class="md-nav__item"> <a href="#component-3-hybrid-scoring-system" class="md-nav__link"> Component 3: Hybrid Scoring System </a> </li> <li class="md-nav__item"> <a href="#component-4-natural-activation-api" class="md-nav__link"> Component 4: Natural Activation API </a> </li> </ul> </nav> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#implementation-phases" class="md-nav__link"> Implementation Phases </a> <nav class="md-nav" aria-label="Implementation Phases"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#phase-1-core-spreading-activation-3-weeks" class="md-nav__link"> Phase 1: Core Spreading Activation (3 weeks) </a> </li> <li class="md-nav__item"> <a href="#phase-2-mcp-tool-integration-2-weeks" class="md-nav__link"> Phase 2: MCP Tool Integration (2 weeks) </a> </li> <li class="md-nav__item"> <a href="#phase-3-advanced-features-3-weeks" class="md-nav__link"> Phase 3: Advanced Features (3 weeks) </a> </li> <li class="md-nav__item"> <a href="#phase-4-production-tuning-2-weeks" class="md-nav__link"> Phase 4: Production Tuning (2 weeks) </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#integration-points" class="md-nav__link"> Integration Points </a> <nav class="md-nav" aria-label="Integration Points"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#1-mcp-server-entry-point" class="md-nav__link"> 1. MCP Server Entry Point </a> </li> <li class="md-nav__item"> <a href="#2-integration-with-existing-tools" class="md-nav__link"> 2. Integration with Existing Tools </a> </li> <li class="md-nav__item"> <a href="#3-conversational-activation-integration" class="md-nav__link"> 3. Conversational Activation Integration </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#success-metrics" class="md-nav__link"> Success Metrics </a> <nav class="md-nav" aria-label="Success Metrics"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#quantitative-metrics" class="md-nav__link"> Quantitative Metrics </a> </li> <li class="md-nav__item"> <a href="#qualitative-metrics" class="md-nav__link"> Qualitative Metrics </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#configuration" class="md-nav__link"> Configuration </a> </li> <li class="md-nav__item"> <a href="#dependencies" class="md-nav__link"> Dependencies </a> <nav class="md-nav" aria-label="Dependencies"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#python-packages" class="md-nav__link"> Python Packages </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#future-enhancements" class="md-nav__link"> Future Enhancements </a> <nav class="md-nav" aria-label="Future Enhancements"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#short-term-next-6-months" class="md-nav__link"> Short-Term (Next 6 Months) </a> </li> <li class="md-nav__item"> <a href="#medium-term-6-12-months" class="md-nav__link"> Medium-Term (6-12 Months) </a> </li> <li class="md-nav__item"> <a href="#long-term-12-months" class="md-nav__link"> Long-Term (12+ Months) </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#risks-mitigations" class="md-nav__link"> Risks & Mitigations </a> <nav class="md-nav" aria-label="Risks & Mitigations"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#risk-1-over-activation-too-many-memories" class="md-nav__link"> Risk 1: Over-Activation (Too Many Memories) </a> </li> <li class="md-nav__item"> <a href="#risk-2-under-activation-missing-relevant-memories" class="md-nav__link"> Risk 2: Under-Activation (Missing Relevant Memories) </a> </li> <li class="md-nav__item"> <a href="#risk-3-graph-sparsity-insufficient-relations" class="md-nav__link"> Risk 3: Graph Sparsity (Insufficient Relations) </a> </li> <li class="md-nav__item"> <a href="#risk-4-latency-from-graph-traversal" class="md-nav__link"> Risk 4: Latency from Graph Traversal </a> </li> <li class="md-nav__item"> <a href="#risk-5-category-inference-accuracy" class="md-nav__link"> Risk 5: Category Inference Accuracy </a> </li> </ul> </nav> </li> <li class="md-nav__item"> <a href="#timeline-summary" class="md-nav__link"> Timeline Summary </a> </li> <li class="md-nav__item"> <a href="#conclusion" class="md-nav__link"> Conclusion </a> </li> <li class="md-nav__item"> <a href="#references" class="md-nav__link"> References </a> <nav class="md-nav" aria-label="References"> <ul class="md-nav__list"> <li class="md-nav__item"> <a href="#academic-papers-2024-2025" class="md-nav__link"> Academic Papers (2024-2025) </a> </li> <li class="md-nav__item"> <a href="#industry-systems" class="md-nav__link"> Industry Systems </a> </li> <li class="md-nav__item"> <a href="#cortexgraph-documentation" class="md-nav__link"> CortexGraph Documentation </a> </li> </ul> </nav> </li> </ul> </nav> </div> </div> </div> <div class="md-content" data-md-component="content"> <article class="md-content__inner md-typeset"> <a href="https://github.com/prefrontal-systems/cortexgraph/edit/main/docs/natural-language-activation-plan.md" title="Edit this page" class="md-content__button md-icon" rel="edit"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M10 20H6V4h7v5h5v3.1l2-2V8l-6-6H6c-1.1 0-2 .9-2 2v16c0 1.1.9 2 2 2h4zm10.2-7c.1 0 .3.1.4.2l1.3 1.3c.2.2.2.6 0 .8l-1 1-2.1-2.1 1-1c.1-.1.2-.2.4-.2m0 3.9L14.1 23H12v-2.1l6.1-6.1z"/></svg> </a> <h1 id="natural-language-activation-implementation-plan">Natural Language Activation Implementation Plan<a class="headerlink" href="#natural-language-activation-implementation-plan" title="Permanent link">¶</a></h1> <h2 id="spreading-activation-for-cortexgraph-memory-recall">Spreading Activation for CortexGraph Memory Recall<a class="headerlink" href="#spreading-activation-for-cortexgraph-memory-recall" title="Permanent link">¶</a></h2> Document Type: Implementation Plan Created: 2025-11-14 Status: Research Complete, Ready for Implementation Feature Branch: <code>feat/natural-language-activation</code> Target Version: v0.6.0 <hr /> <h2 id="executive-summary">Executive Summary<a class="headerlink" href="#executive-summary" title="Permanent link">¶</a></h2> This document outlines the implementation of natural language activation for cortexgraph, enabling conversational memory recall through spreading activation and multi-dimensional search. Unlike the existing explicit tool-based retrieval (<code>search_memory</code>), this system will automatically activate related memories based on conversational context, creating a more natural and human-like memory experience. Core Innovation: Hybrid spreading activation + temporal decay, combining graph-based memory traversal with cortexgraph's unique temporal properties. Expected Impact: 3-4x improvement in context-relevant memory retrieval during conversations Timeline: 8-10 weeks to production-ready system <hr /> <h2 id="table-of-contents">Table of Contents<a class="headerlink" href="#table-of-contents" title="Permanent link">¶</a></h2> <ol> <li><a href="#problem-statement">Problem Statement</a></li> <li><a href="#research-findings">Research Findings</a></li> <li><a href="#solution-architecture">Solution Architecture</a></li> <li><a href="#implementation-phases">Implementation Phases</a></li> <li><a href="#integration-points">Integration Points</a></li> <li><a href="#success-metrics">Success Metrics</a></li> <li><a href="#future-enhancements">Future Enhancements</a></li> </ol> <hr /> <h2 id="problem-statement">Problem Statement<a class="headerlink" href="#problem-statement" title="Permanent link">¶</a></h2> <h3 id="current-state">Current State<a class="headerlink" href="#current-state" title="Permanent link">¶</a></h3> CortexGraph v0.5.1 provides excellent memory foundations: - ✅ Temporal decay with natural spaced repetition - ✅ Knowledge graph with entities and relations - ✅ Multi-message context (via <code>observe_memory_usage</code>) - ✅ Review priority system (danger zone detection) - ✅ Cross-domain usage detection (Jaccard similarity <30%) However, memory retrieval requires explicit search queries: <div class="highlight"><pre><code><a id="__codelineno-0-1" name="__codelineno-0-1" href="#__codelineno-0-1"></a># Current: Explicit search required <a id="__codelineno-0-2" name="__codelineno-0-2" href="#__codelineno-0-2"></a>search_memory(query="TypeScript preferences", tags=["backend"]) </code></pre></div> <h3 id="the-gap-natural-conversational-activation">The Gap: Natural Conversational Activation<a class="headerlink" href="#the-gap-natural-conversational-activation" title="Permanent link">¶</a></h3> When humans converse, related memories activate automatically without explicit recall commands: Example Conversation: <div class="highlight"><pre><code><a id="__codelineno-1-1" name="__codelineno-1-1" href="#__codelineno-1-1"></a>User: "I'm starting a new backend API project" <a id="__codelineno-1-2" name="__codelineno-1-2" href="#__codelineno-1-2"></a>→ Should auto-activate: <a id="__codelineno-1-3" name="__codelineno-1-3" href="#__codelineno-1-3"></a> - Previous preference for TypeScript <a id="__codelineno-1-4" name="__codelineno-1-4" href="#__codelineno-1-4"></a> - Recent discussion about PostgreSQL for analytics <a id="__codelineno-1-5" name="__codelineno-1-5" href="#__codelineno-1-5"></a> - Decision to use JWT for authentication <a id="__codelineno-1-6" name="__codelineno-1-6" href="#__codelineno-1-6"></a> - Related project X architecture notes </code></pre></div> Current behavior: LLM must explicitly decide to search for each relevant memory Desired behavior: Related memories surface automatically through spreading activation <h3 id="why-this-matters">Why This Matters<a class="headerlink" href="#why-this-matters" title="Permanent link">¶</a></h3> From User Perspective: - AI remembers context without being prompted - Feels more natural and attentive - Reduces cognitive load (user doesn't need to remind AI) From System Perspective: - Leverages existing knowledge graph structure - Complements (doesn't replace) explicit search - Aligns with cortexgraph's temporal memory philosophy <hr /> <h2 id="research-findings">Research Findings<a class="headerlink" href="#research-findings" title="Permanent link">¶</a></h2> <h3 id="state-of-the-art-2024-2025">State-of-the-Art (2024-2025)<a class="headerlink" href="#state-of-the-art-2024-2025" title="Permanent link">¶</a></h3> <h4 id="1-imdmr-multi-dimensional-memory-retrieval-nov-2025">1. IMDMR: Multi-Dimensional Memory Retrieval (Nov 2025)<a class="headerlink" href="#1-imdmr-multi-dimensional-memory-retrieval-nov-2025" title="Permanent link">¶</a></h4> Paper: arxiv:2511.05495v1 Key Finding: 3.8x improvement using 6-dimensional search Six Dimensions: 1. Semantic - meaning similarity (embeddings) 2. Entity - shared named entities 3. Category - topical classification 4. Intent - user goals/preferences 5. Context - conversational state 6. Temporal - time-based relevance Relevance to CortexGraph: - ✅ Already have: Semantic (optional embeddings), Entity (graph), Temporal (decay) - ❌ Need: Category, Intent, Context dimensions Performance: Individual dimensions vs. full system = 23.3% improvement <hr /> <h4 id="2-spreadpy-spreading-activation-library-july-2025">2. SpreadPy: Spreading Activation Library (July 2025)<a class="headerlink" href="#2-spreadpy-spreading-activation-library-july-2025" title="Permanent link">¶</a></h4> Paper: arxiv:2507.09628 GitHub: Python library for cognitive network activation Core Algorithm: <div class="highlight"><pre><code><a id="__codelineno-2-1" name="__codelineno-2-1" href="#__codelineno-2-1"></a># Spreading activation pseudocode <a id="__codelineno-2-2" name="__codelineno-2-2" href="#__codelineno-2-2"></a>def spread_activation(source_nodes, network, decay_rate, threshold): <a id="__codelineno-2-3" name="__codelineno-2-3" href="#__codelineno-2-3"></a> """ <a id="__codelineno-2-4" name="__codelineno-2-4" href="#__codelineno-2-4"></a> Args: <a id="__codelineno-2-5" name="__codelineno-2-5" href="#__codelineno-2-5"></a> source_nodes: Initial activation points (e.g., entities in user message) <a id="__codelineno-2-6" name="__codelineno-2-6" href="#__codelineno-2-6"></a> network: Graph structure (nodes = memories, edges = relations) <a id="__codelineno-2-7" name="__codelineno-2-7" href="#__codelineno-2-7"></a> decay_rate: Activation strength decay per hop <a id="__codelineno-2-8" name="__codelineno-2-8" href="#__codelineno-2-8"></a> threshold: Minimum activation to consider node "activated" <a id="__codelineno-2-9" name="__codelineno-2-9" href="#__codelineno-2-9"></a> <a id="__codelineno-2-10" name="__codelineno-2-10" href="#__codelineno-2-10"></a> Returns: <a id="__codelineno-2-11" name="__codelineno-2-11" href="#__codelineno-2-11"></a> activated_nodes: Memories that received sufficient activation <a id="__codelineno-2-12" name="__codelineno-2-12" href="#__codelineno-2-12"></a> """ <a id="__codelineno-2-13" name="__codelineno-2-13" href="#__codelineno-2-13"></a> activation = {node: 0.0 for node in network.nodes} <a id="__codelineno-2-14" name="__codelineno-2-14" href="#__codelineno-2-14"></a> <a id="__codelineno-2-15" name="__codelineno-2-15" href="#__codelineno-2-15"></a> # Initialize source nodes <a id="__codelineno-2-16" name="__codelineno-2-16" href="#__codelineno-2-16"></a> for source in source_nodes: <a id="__codelineno-2-17" name="__codelineno-2-17" href="#__codelineno-2-17"></a> activation[source] = 1.0 <a id="__codelineno-2-18" name="__codelineno-2-18" href="#__codelineno-2-18"></a> <a id="__codelineno-2-19" name="__codelineno-2-19" href="#__codelineno-2-19"></a> # Spread activation iteratively <a id="__codelineno-2-20" name="__codelineno-2-20" href="#__codelineno-2-20"></a> for iteration in range(max_hops): <a id="__codelineno-2-21" name="__codelineno-2-21" href="#__codelineno-2-21"></a> new_activation = activation.copy() <a id="__codelineno-2-22" name="__codelineno-2-22" href="#__codelineno-2-22"></a> <a id="__codelineno-2-23" name="__codelineno-2-23" href="#__codelineno-2-23"></a> for node, strength in activation.items(): <a id="__codelineno-2-24" name="__codelineno-2-24" href="#__codelineno-2-24"></a> if strength < threshold: <a id="__codelineno-2-25" name="__codelineno-2-25" href="#__codelineno-2-25"></a> continue <a id="__codelineno-2-26" name="__codelineno-2-26" href="#__codelineno-2-26"></a> <a id="__codelineno-2-27" name="__codelineno-2-27" href="#__codelineno-2-27"></a> # Spread to neighbors <a id="__codelineno-2-28" name="__codelineno-2-28" href="#__codelineno-2-28"></a> for neighbor in network.neighbors(node): <a id="__codelineno-2-29" name="__codelineno-2-29" href="#__codelineno-2-29"></a> edge_weight = network[node][neighbor]['weight'] <a id="__codelineno-2-30" name="__codelineno-2-30" href="#__codelineno-2-30"></a> transferred = strength * edge_weight * (decay_rate ** iteration) <a id="__codelineno-2-31" name="__codelineno-2-31" href="#__codelineno-2-31"></a> new_activation[neighbor] += transferred <a id="__codelineno-2-32" name="__codelineno-2-32" href="#__codelineno-2-32"></a> <a id="__codelineno-2-33" name="__codelineno-2-33" href="#__codelineno-2-33"></a> activation = new_activation <a id="__codelineno-2-34" name="__codelineno-2-34" href="#__codelineno-2-34"></a> <a id="__codelineno-2-35" name="__codelineno-2-35" href="#__codelineno-2-35"></a> # Return nodes above threshold <a id="__codelineno-2-36" name="__codelineno-2-36" href="#__codelineno-2-36"></a> return {n: a for n, a in activation.items() if a >= threshold} </code></pre></div> Relevance to CortexGraph: - Direct application to existing knowledge graph (entities + relations) - Compatible with temporal decay (combine activation strength with decay score) - Can use existing relation weights (strength field) <hr /> <h4 id="3-mem0-scalable-production-memory-apr-2025">3. Mem0: Scalable Production Memory (Apr 2025)<a class="headerlink" href="#3-mem0-scalable-production-memory-apr-2025" title="Permanent link">¶</a></h4> Paper: arxiv:2504.19413 Key Finding: 26% improvement over OpenAI, 91% lower latency Architecture: <div class="highlight"><pre><code><a id="__codelineno-3-1" name="__codelineno-3-1" href="#__codelineno-3-1"></a>User Message <a id="__codelineno-3-2" name="__codelineno-3-2" href="#__codelineno-3-2"></a> ↓ <a id="__codelineno-3-3" name="__codelineno-3-3" href="#__codelineno-3-3"></a>Extract Facts (LLM) <a id="__codelineno-3-4" name="__codelineno-3-4" href="#__codelineno-3-4"></a> ↓ <a id="__codelineno-3-5" name="__codelineno-3-5" href="#__codelineno-3-5"></a>Update Memory Graph (deduplicate, consolidate) <a id="__codelineno-3-6" name="__codelineno-3-6" href="#__codelineno-3-6"></a> ↓ <a id="__codelineno-3-7" name="__codelineno-3-7" href="#__codelineno-3-7"></a>Retrieve Relevant Context (RAG + Graph) </code></pre></div> Mem0ᵍ Enhancement: Graph-based store for multi-session relationships Relevance to CortexGraph: - Validates graph-enhanced memory approach - Two-phase pipeline: Extract → Update (aligns with conversational-activation-plan.md) - Confirms value of deduplication (already in cortexgraph roadmap) <hr /> <h4 id="4-a-mem-agentic-memory-with-zettelkasten-feb-2025">4. A-MEM: Agentic Memory with Zettelkasten (Feb 2025)<a class="headerlink" href="#4-a-mem-agentic-memory-with-zettelkasten-feb-2025" title="Permanent link">¶</a></h4> Paper: arxiv:2502.12110 Key Concept: Dynamic indexing with interconnected notes Memory Structure: <div class="highlight"><pre><code><a id="__codelineno-4-1" name="__codelineno-4-1" href="#__codelineno-4-1"></a>{ <a id="__codelineno-4-2" name="__codelineno-4-2" href="#__codelineno-4-2"></a> "content": "User prefers TypeScript for backend", <a id="__codelineno-4-3" name="__codelineno-4-3" href="#__codelineno-4-3"></a> "keywords": ["typescript", "backend", "preference"], <a id="__codelineno-4-4" name="__codelineno-4-4" href="#__codelineno-4-4"></a> "tags": ["programming", "languages"], <a id="__codelineno-4-5" name="__codelineno-4-5" href="#__codelineno-4-5"></a> "links": [ <a id="__codelineno-4-6" name="__codelineno-4-6" href="#__codelineno-4-6"></a> {"to": "mem-456", "relation": "related_to", "context": "same project"}, <a id="__codelineno-4-7" name="__codelineno-4-7" href="#__codelineno-4-7"></a> {"to": "mem-789", "relation": "elaborates_on", "context": "tech stack"} <a id="__codelineno-4-8" name="__codelineno-4-8" href="#__codelineno-4-8"></a> ], <a id="__codelineno-4-9" name="__codelineno-4-9" href="#__codelineno-4-9"></a> "context": "Discussion about new API project" <a id="__codelineno-4-10" name="__codelineno-4-10" href="#__codelineno-4-10"></a>} </code></pre></div> Relevance to CortexGraph: - ✅ Already have: content, tags, entities - ✅ Already have: relations (via <code>create_relation</code> tool) - ❌ Need: Automatic keyword extraction - ❌ Need: Contextual link creation <hr /> <h4 id="5-context-window-paradox-industry-research-2025">5. Context Window Paradox (Industry Research 2025)<a class="headerlink" href="#5-context-window-paradox-industry-research-2025" title="Permanent link">¶</a></h4> Finding: Beyond 128K tokens, LLM performance degrades ("context rot") Implication: Active memory retrieval > dumping entire context Solution: Intelligent activation that surfaces only relevant memories Relevance to CortexGraph: - Validates selective memory retrieval approach - Spreading activation naturally limits context to relevant memories - Temporal decay filters out stale information <hr /> <h3 id="synthesis-what-cortexgraph-needs">Synthesis: What CortexGraph Needs<a class="headerlink" href="#synthesis-what-cortexgraph-needs" title="Permanent link">¶</a></h3> Combining all research findings: <table> <thead> <tr> <th>Feature</th> <th>IMDMR</th> <th>SpreadPy</th> <th>Mem0</th> <th>A-MEM</th> <th>CortexGraph Status</th> </tr> </thead> <tbody> <tr> <td>Semantic Search</td> <td>✅</td> <td></td> <td>✅</td> <td></td> <td>✅ (optional embeddings)</td> </tr> <tr> <td>Entity Tracking</td> <td>✅</td> <td></td> <td>✅</td> <td>✅</td> <td>✅ (graph entities)</td> </tr> <tr> <td>Temporal Relevance</td> <td>✅</td> <td></td> <td></td> <td></td> <td>✅ (decay + review priority)</td> </tr> <tr> <td>Spreading Activation</td> <td></td> <td>✅</td> <td></td> <td></td> <td>❌ NEED</td> </tr> <tr> <td>Category/Intent</td> <td>✅</td> <td></td> <td></td> <td></td> <td>❌ NEED</td> </tr> <tr> <td>Context Dimension</td> <td>✅</td> <td></td> <td></td> <td>✅</td> <td>⚠️ Partial (tags only)</td> </tr> <tr> <td>Automatic Activation</td> <td>✅</td> <td>✅</td> <td>✅</td> <td></td> <td>❌ NEED</td> </tr> <tr> <td>Dynamic Relations</td> <td></td> <td></td> <td>✅</td> <td>✅</td> <td>⚠️ Manual only</td> </tr> </tbody> </table> Priority Gaps: 1. Spreading activation engine - Core algorithm for graph traversal 2. Automatic activation triggers - Detect when to activate vs. explicit search 3. Context extraction - Pull entities/intents from conversation 4. Category inference - Classify memory topical areas <hr /> <h2 id="solution-architecture">Solution Architecture<a class="headerlink" href="#solution-architecture" title="Permanent link">¶</a></h2> <h3 id="three-layer-activation-system">Three-Layer Activation System<a class="headerlink" href="#three-layer-activation-system" title="Permanent link">¶</a></h3> <div class="highlight"><pre><code><a id="__codelineno-5-1" name="__codelineno-5-1" href="#__codelineno-5-1"></a>┌─────────────────────────────────────────────────────────┐ <a id="__codelineno-5-2" name="__codelineno-5-2" href="#__codelineno-5-2"></a>│ Layer 1: Trigger Detection │ <a id="__codelineno-5-3" name="__codelineno-5-3" href="#__codelineno-5-3"></a>│ - Extract entities from user message │ <a id="__codelineno-5-4" name="__codelineno-5-4" href="#__codelineno-5-4"></a>│ - Detect intent (question, statement, command) │ <a id="__codelineno-5-5" name="__codelineno-5-5" href="#__codelineno-5-5"></a>│ - Determine activation vs. explicit search │ <a id="__codelineno-5-6" name="__codelineno-5-6" href="#__codelineno-5-6"></a>└────────────────────┬────────────────────────────────────┘ <a id="__codelineno-5-7" name="__codelineno-5-7" href="#__codelineno-5-7"></a> │ <a id="__codelineno-5-8" name="__codelineno-5-8" href="#__codelineno-5-8"></a>┌────────────────────▼────────────────────────────────────┐ <a id="__codelineno-5-9" name="__codelineno-5-9" href="#__codelineno-5-9"></a>│ Layer 2: Spreading Activation Engine │ <a id="__codelineno-5-10" name="__codelineno-5-10" href="#__codelineno-5-10"></a>│ - Initialize activation from source entities │ <a id="__codelineno-5-11" name="__codelineno-5-11" href="#__codelineno-5-11"></a>│ - Propagate through relation graph │ <a id="__codelineno-5-12" name="__codelineno-5-12" href="#__codelineno-5-12"></a>│ - Combine with temporal decay scores │ <a id="__codelineno-5-13" name="__codelineno-5-13" href="#__codelineno-5-13"></a>│ - Apply cross-domain detection │ <a id="__codelineno-5-14" name="__codelineno-5-14" href="#__codelineno-5-14"></a>└────────────────────┬────────────────────────────────────┘ <a id="__codelineno-5-15" name="__codelineno-5-15" href="#__codelineno-5-15"></a> │ <a id="__codelineno-5-16" name="__codelineno-5-16" href="#__codelineno-5-16"></a>┌────────────────────▼────────────────────────────────────┐ <a id="__codelineno-5-17" name="__codelineno-5-17" href="#__codelineno-5-17"></a>│ Layer 3: Memory Integration │ <a id="__codelineno-5-18" name="__codelineno-5-18" href="#__codelineno-5-18"></a>│ - Blend activated memories with review candidates │ <a id="__codelineno-5-19" name="__codelineno-5-19" href="#__codelineno-5-19"></a>│ - Rank by combined score (activation × decay × review) │ <a id="__codelineno-5-20" name="__codelineno-5-20" href="#__codelineno-5-20"></a>│ - Return top-k for LLM context │ <a id="__codelineno-5-21" name="__codelineno-5-21" href="#__codelineno-5-21"></a>│ - Call observe_memory_usage for reinforcement │ <a id="__codelineno-5-22" name="__codelineno-5-22" href="#__codelineno-5-22"></a>└─────────────────────────────────────────────────────────┘ </code></pre></div> <h3 id="core-components">Core Components<a class="headerlink" href="#core-components" title="Permanent link">¶</a></h3> <h4 id="component-1-activation-trigger-detector">Component 1: Activation Trigger Detector<a class="headerlink" href="#component-1-activation-trigger-detector" title="Permanent link">¶</a></h4> Purpose: Determine when to activate memories automatically vs. wait for explicit search Implementation: <div class="highlight"><pre><code><a id="__codelineno-6-1" name="__codelineno-6-1" href="#__codelineno-6-1"></a># src/cortexgraph/activation/trigger_detector.py <a id="__codelineno-6-2" name="__codelineno-6-2" href="#__codelineno-6-2"></a> <a id="__codelineno-6-3" name="__codelineno-6-3" href="#__codelineno-6-3"></a>from typing import List, Dict, Literal <a id="__codelineno-6-4" name="__codelineno-6-4" href="#__codelineno-6-4"></a>import spacy <a id="__codelineno-6-5" name="__codelineno-6-5" href="#__codelineno-6-5"></a> <a id="__codelineno-6-6" name="__codelineno-6-6" href="#__codelineno-6-6"></a>class ActivationTrigger: <a id="__codelineno-6-7" name="__codelineno-6-7" href="#__codelineno-6-7"></a> def __init__(self): <a id="__codelineno-6-8" name="__codelineno-6-8" href="#__codelineno-6-8"></a> self.nlp = spacy.load("en_core_web_sm") <a id="__codelineno-6-9" name="__codelineno-6-9" href="#__codelineno-6-9"></a> self.question_patterns = ["what", "when", "where", "who", "how", "why", "do you remember"] <a id="__codelineno-6-10" name="__codelineno-6-10" href="#__codelineno-6-10"></a> <a id="__codelineno-6-11" name="__codelineno-6-11" href="#__codelineno-6-11"></a> def detect(self, message: str) -> Dict: <a id="__codelineno-6-12" name="__codelineno-6-12" href="#__codelineno-6-12"></a> """ <a id="__codelineno-6-13" name="__codelineno-6-13" href="#__codelineno-6-13"></a> Determine if message should trigger automatic activation. <a id="__codelineno-6-14" name="__codelineno-6-14" href="#__codelineno-6-14"></a> <a id="__codelineno-6-15" name="__codelineno-6-15" href="#__codelineno-6-15"></a> Returns: <a id="__codelineno-6-16" name="__codelineno-6-16" href="#__codelineno-6-16"></a> { <a id="__codelineno-6-17" name="__codelineno-6-17" href="#__codelineno-6-17"></a> "should_activate": bool, <a id="__codelineno-6-18" name="__codelineno-6-18" href="#__codelineno-6-18"></a> "activation_type": "question" | "statement" | "command", <a id="__codelineno-6-19" name="__codelineno-6-19" href="#__codelineno-6-19"></a> "source_entities": List[str], # Entities to start spreading from <a id="__codelineno-6-20" name="__codelineno-6-20" href="#__codelineno-6-20"></a> "intent": "recall" | "store" | "update" | "general" <a id="__codelineno-6-21" name="__codelineno-6-21" href="#__codelineno-6-21"></a> } <a id="__codelineno-6-22" name="__codelineno-6-22" href="#__codelineno-6-22"></a> """ <a id="__codelineno-6-23" name="__codelineno-6-23" href="#__codelineno-6-23"></a> doc = self.nlp(message) <a id="__codelineno-6-24" name="__codelineno-6-24" href="#__codelineno-6-24"></a> <a id="__codelineno-6-25" name="__codelineno-6-25" href="#__codelineno-6-25"></a> # Extract entities <a id="__codelineno-6-26" name="__codelineno-6-26" href="#__codelineno-6-26"></a> entities = [ent.text.lower() for ent in doc.ents] <a id="__codelineno-6-27" name="__codelineno-6-27" href="#__codelineno-6-27"></a> <a id="__codelineno-6-28" name="__codelineno-6-28" href="#__codelineno-6-28"></a> # Detect question (triggers recall activation) <a id="__codelineno-6-29" name="__codelineno-6-29" href="#__codelineno-6-29"></a> is_question = any(token.text.lower() in self.question_patterns for token in doc) <a id="__codelineno-6-30" name="__codelineno-6-30" href="#__codelineno-6-30"></a> <a id="__codelineno-6-31" name="__codelineno-6-31" href="#__codelineno-6-31"></a> # Detect explicit memory command <a id="__codelineno-6-32" name="__codelineno-6-32" href="#__codelineno-6-32"></a> memory_keywords = ["remember", "recall", "what did i say", "previously"] <a id="__codelineno-6-33" name="__codelineno-6-33" href="#__codelineno-6-33"></a> is_memory_command = any(kw in message.lower() for kw in memory_keywords) <a id="__codelineno-6-34" name="__codelineno-6-34" href="#__codelineno-6-34"></a> <a id="__codelineno-6-35" name="__codelineno-6-35" href="#__codelineno-6-35"></a> # Intent classification <a id="__codelineno-6-36" name="__codelineno-6-36" href="#__codelineno-6-36"></a> if is_question or is_memory_command: <a id="__codelineno-6-37" name="__codelineno-6-37" href="#__codelineno-6-37"></a> intent = "recall" <a id="__codelineno-6-38" name="__codelineno-6-38" href="#__codelineno-6-38"></a> should_activate = len(entities) > 0 # Activate if entities present <a id="__codelineno-6-39" name="__codelineno-6-39" href="#__codelineno-6-39"></a> elif any(token.pos_ == "VERB" and token.lemma_ in ["prefer", "like", "decide", "choose"] for token in doc): <a id="__codelineno-6-40" name="__codelineno-6-40" href="#__codelineno-6-40"></a> intent = "store" # Preference/decision statement <a id="__codelineno-6-41" name="__codelineno-6-41" href="#__codelineno-6-41"></a> should_activate = False # Don't activate on save <a id="__codelineno-6-42" name="__codelineno-6-42" href="#__codelineno-6-42"></a> else: <a id="__codelineno-6-43" name="__codelineno-6-43" href="#__codelineno-6-43"></a> intent = "general" <a id="__codelineno-6-44" name="__codelineno-6-44" href="#__codelineno-6-44"></a> should_activate = len(entities) >= 2 # Activate if multiple entities (likely building on prior context) <a id="__codelineno-6-45" name="__codelineno-6-45" href="#__codelineno-6-45"></a> <a id="__codelineno-6-46" name="__codelineno-6-46" href="#__codelineno-6-46"></a> return { <a id="__codelineno-6-47" name="__codelineno-6-47" href="#__codelineno-6-47"></a> "should_activate": should_activate, <a id="__codelineno-6-48" name="__codelineno-6-48" href="#__codelineno-6-48"></a> "activation_type": "question" if is_question else "statement", <a id="__codelineno-6-49" name="__codelineno-6-49" href="#__codelineno-6-49"></a> "source_entities": entities, <a id="__codelineno-6-50" name="__codelineno-6-50" href="#__codelineno-6-50"></a> "intent": intent <a id="__codelineno-6-51" name="__codelineno-6-51" href="#__codelineno-6-51"></a> } </code></pre></div> Test Coverage: - Detect questions correctly (95%+ accuracy) - Extract entities from messages (spaCy NER) - Intent classification (recall vs. store vs. general) <hr /> <h4 id="component-2-spreading-activation-engine">Component 2: Spreading Activation Engine<a class="headerlink" href="#component-2-spreading-activation-engine" title="Permanent link">¶</a></h4> Purpose: Traverse knowledge graph from source entities, activating related memories Algorithm: Multi-hop activation with temporal decay integration Implementation: <div class="highlight"><pre><code><a id="__codelineno-7-1" name="__codelineno-7-1" href="#__codelineno-7-1"></a># src/cortexgraph/activation/spreading.py <a id="__codelineno-7-2" name="__codelineno-7-2" href="#__codelineno-7-2"></a> <a id="__codelineno-7-3" name="__codelineno-7-3" href="#__codelineno-7-3"></a>from typing import List, Dict, Set <a id="__codelineno-7-4" name="__codelineno-7-4" href="#__codelineno-7-4"></a>from collections import defaultdict <a id="__codelineno-7-5" name="__codelineno-7-5" href="#__codelineno-7-5"></a>import networkx as nx <a id="__codelineno-7-6" name="__codelineno-7-6" href="#__codelineno-7-6"></a> <a id="__codelineno-7-7" name="__codelineno-7-7" href="#__codelineno-7-7"></a>class SpreadingActivation: <a id="__codelineno-7-8" name="__codelineno-7-8" href="#__codelineno-7-8"></a> def __init__(self, storage, config): <a id="__codelineno-7-9" name="__codelineno-7-9" href="#__codelineno-7-9"></a> self.storage = storage <a id="__codelineno-7-10" name="__codelineno-7-10" href="#__codelineno-7-10"></a> self.decay_rate = config.ACTIVATION_DECAY_RATE # 0.7 default <a id="__codelineno-7-11" name="__codelineno-7-11" href="#__codelineno-7-11"></a> self.threshold = config.ACTIVATION_THRESHOLD # 0.15 default <a id="__codelineno-7-12" name="__codelineno-7-12" href="#__codelineno-7-12"></a> self.max_hops = config.MAX_ACTIVATION_HOPS # 3 default <a id="__codelineno-7-13" name="__codelineno-7-13" href="#__codelineno-7-13"></a> <a id="__codelineno-7-14" name="__codelineno-7-14" href="#__codelineno-7-14"></a> def activate(self, source_entities: List[str]) -> Dict[str, float]: <a id="__codelineno-7-15" name="__codelineno-7-15" href="#__codelineno-7-15"></a> """ <a id="__codelineno-7-16" name="__codelineno-7-16" href="#__codelineno-7-16"></a> Spread activation from source entities through knowledge graph. <a id="__codelineno-7-17" name="__codelineno-7-17" href="#__codelineno-7-17"></a> <a id="__codelineno-7-18" name="__codelineno-7-18" href="#__codelineno-7-18"></a> Args: <a id="__codelineno-7-19" name="__codelineno-7-19" href="#__codelineno-7-19"></a> source_entities: List of entity names to start activation <a id="__codelineno-7-20" name="__codelineno-7-20" href="#__codelineno-7-20"></a> <a id="__codelineno-7-21" name="__codelineno-7-21" href="#__codelineno-7-21"></a> Returns: <a id="__codelineno-7-22" name="__codelineno-7-22" href="#__codelineno-7-22"></a> activated_memories: {memory_id: activation_score} <a id="__codelineno-7-23" name="__codelineno-7-23" href="#__codelineno-7-23"></a> """ <a id="__codelineno-7-24" name="__codelineno-7-24" href="#__codelineno-7-24"></a> # Build activation graph from memory relations <a id="__codelineno-7-25" name="__codelineno-7-25" href="#__codelineno-7-25"></a> graph = self._build_activation_graph() <a id="__codelineno-7-26" name="__codelineno-7-26" href="#__codelineno-7-26"></a> <a id="__codelineno-7-27" name="__codelineno-7-27" href="#__codelineno-7-27"></a> # Initialize activation <a id="__codelineno-7-28" name="__codelineno-7-28" href="#__codelineno-7-28"></a> activation = defaultdict(float) <a id="__codelineno-7-29" name="__codelineno-7-29" href="#__codelineno-7-29"></a> for entity in source_entities: <a id="__codelineno-7-30" name="__codelineno-7-30" href="#__codelineno-7-30"></a> # Find memories containing this entity <a id="__codelineno-7-31" name="__codelineno-7-31" href="#__codelineno-7-31"></a> memory_ids = self._find_memories_by_entity(entity) <a id="__codelineno-7-32" name="__codelineno-7-32" href="#__codelineno-7-32"></a> for mid in memory_ids: <a id="__codelineno-7-33" name="__codelineno-7-33" href="#__codelineno-7-33"></a> activation[mid] = 1.0 <a id="__codelineno-7-34" name="__codelineno-7-34" href="#__codelineno-7-34"></a> <a id="__codelineno-7-35" name="__codelineno-7-35" href="#__codelineno-7-35"></a> if not activation: <a id="__codelineno-7-36" name="__codelineno-7-36" href="#__codelineno-7-36"></a> return {} <a id="__codelineno-7-37" name="__codelineno-7-37" href="#__codelineno-7-37"></a> <a id="__codelineno-7-38" name="__codelineno-7-38" href="#__codelineno-7-38"></a> # Spread activation iteratively <a id="__codelineno-7-39" name="__codelineno-7-39" href="#__codelineno-7-39"></a> for hop in range(self.max_hops): <a id="__codelineno-7-40" name="__codelineno-7-40" href="#__codelineno-7-40"></a> new_activation = activation.copy() <a id="__codelineno-7-41" name="__codelineno-7-41" href="#__codelineno-7-41"></a> current_decay = self.decay_rate ** (hop + 1) <a id="__codelineno-7-42" name="__codelineno-7-42" href="#__codelineno-7-42"></a> <a id="__codelineno-7-43" name="__codelineno-7-43" href="#__codelineno-7-43"></a> for memory_id, strength in activation.items(): <a id="__codelineno-7-44" name="__codelineno-7-44" href="#__codelineno-7-44"></a> if strength < self.threshold: <a id="__codelineno-7-45" name="__codelineno-7-45" href="#__codelineno-7-45"></a> continue <a id="__codelineno-7-46" name="__codelineno-7-46" href="#__codelineno-7-46"></a> <a id="__codelineno-7-47" name="__codelineno-7-47" href="#__codelineno-7-47"></a> # Get outgoing relations <a id="__codelineno-7-48" name="__codelineno-7-48" href="#__codelineno-7-48"></a> relations = self.storage.get_relations_from(memory_id) <a id="__codelineno-7-49" name="__codelineno-7-49" href="#__codelineno-7-49"></a> <a id="__codelineno-7-50" name="__codelineno-7-50" href="#__codelineno-7-50"></a> for relation in relations: <a id="__codelineno-7-51" name="__codelineno-7-51" href="#__codelineno-7-51"></a> target_id = relation.to_memory_id <a id="__codelineno-7-52" name="__codelineno-7-52" href="#__codelineno-7-52"></a> edge_weight = relation.strength # Use relation strength as edge weight <a id="__codelineno-7-53" name="__codelineno-7-53" href="#__codelineno-7-53"></a> <a id="__codelineno-7-54" name="__codelineno-7-54" href="#__codelineno-7-54"></a> # Transfer activation with decay <a id="__codelineno-7-55" name="__codelineno-7-55" href="#__codelineno-7-55"></a> transferred = strength * edge_weight * current_decay <a id="__codelineno-7-56" name="__codelineno-7-56" href="#__codelineno-7-56"></a> new_activation[target_id] += transferred <a id="__codelineno-7-57" name="__codelineno-7-57" href="#__codelineno-7-57"></a> <a id="__codelineno-7-58" name="__codelineno-7-58" href="#__codelineno-7-58"></a> activation = new_activation <a id="__codelineno-7-59" name="__codelineno-7-59" href="#__codelineno-7-59"></a> <a id="__codelineno-7-60" name="__codelineno-7-60" href="#__codelineno-7-60"></a> # Filter by threshold <a id="__codelineno-7-61" name="__codelineno-7-61" href="#__codelineno-7-61"></a> return {mid: score for mid, score in activation.items() if score >= self.threshold} <a id="__codelineno-7-62" name="__codelineno-7-62" href="#__codelineno-7-62"></a> <a id="__codelineno-7-63" name="__codelineno-7-63" href="#__codelineno-7-63"></a> def _build_activation_graph(self) -> nx.DiGraph: <a id="__codelineno-7-64" name="__codelineno-7-64" href="#__codelineno-7-64"></a> """Build NetworkX graph from memory relations.""" <a id="__codelineno-7-65" name="__codelineno-7-65" href="#__codelineno-7-65"></a> G = nx.DiGraph() <a id="__codelineno-7-66" name="__codelineno-7-66" href="#__codelineno-7-66"></a> relations = self.storage.get_all_relations() <a id="__codelineno-7-67" name="__codelineno-7-67" href="#__codelineno-7-67"></a> <a id="__codelineno-7-68" name="__codelineno-7-68" href="#__codelineno-7-68"></a> for rel in relations: <a id="__codelineno-7-69" name="__codelineno-7-69" href="#__codelineno-7-69"></a> G.add_edge( <a id="__codelineno-7-70" name="__codelineno-7-70" href="#__codelineno-7-70"></a> rel.from_memory_id, <a id="__codelineno-7-71" name="__codelineno-7-71" href="#__codelineno-7-71"></a> rel.to_memory_id, <a id="__codelineno-7-72" name="__codelineno-7-72" href="#__codelineno-7-72"></a> weight=rel.strength, <a id="__codelineno-7-73" name="__codelineno-7-73" href="#__codelineno-7-73"></a> type=rel.relation_type <a id="__codelineno-7-74" name="__codelineno-7-74" href="#__codelineno-7-74"></a> ) <a id="__codelineno-7-75" name="__codelineno-7-75" href="#__codelineno-7-75"></a> <a id="__codelineno-7-76" name="__codelineno-7-76" href="#__codelineno-7-76"></a> return G <a id="__codelineno-7-77" name="__codelineno-7-77" href="#__codelineno-7-77"></a> <a id="__codelineno-7-78" name="__codelineno-7-78" href="#__codelineno-7-78"></a> def _find_memories_by_entity(self, entity: str) -> List[str]: <a id="__codelineno-7-79" name="__codelineno-7-79" href="#__codelineno-7-79"></a> """Find all memory IDs containing given entity.""" <a id="__codelineno-7-80" name="__codelineno-7-80" href="#__codelineno-7-80"></a> all_memories = self.storage.get_all_memories(status="active") <a id="__codelineno-7-81" name="__codelineno-7-81" href="#__codelineno-7-81"></a> return [m.id for m in all_memories if entity.lower() in [e.lower() for e in m.entities]] </code></pre></div> Configuration: <div class="highlight"><pre><code><a id="__codelineno-8-1" name="__codelineno-8-1" href="#__codelineno-8-1"></a># Spreading Activation <a id="__codelineno-8-2" name="__codelineno-8-2" href="#__codelineno-8-2"></a>ACTIVATION_DECAY_RATE=0.7 # Activation strength per hop (70% retained) <a id="__codelineno-8-3" name="__codelineno-8-3" href="#__codelineno-8-3"></a>ACTIVATION_THRESHOLD=0.15 # Minimum activation to consider <a id="__codelineno-8-4" name="__codelineno-8-4" href="#__codelineno-8-4"></a>MAX_ACTIVATION_HOPS=3 # Maximum graph traversal depth </code></pre></div> <hr /> <h4 id="component-3-hybrid-scoring-system">Component 3: Hybrid Scoring System<a class="headerlink" href="#component-3-hybrid-scoring-system" title="Permanent link">¶</a></h4> Purpose: Combine spreading activation with existing temporal decay and review priority Formula: <div class="highlight"><pre><code><a id="__codelineno-9-1" name="__codelineno-9-1" href="#__codelineno-9-1"></a>final_score = activation_score × decay_score × (1 + review_priority) <a id="__codelineno-9-2" name="__codelineno-9-2" href="#__codelineno-9-2"></a> <a id="__codelineno-9-3" name="__codelineno-9-3" href="#__codelineno-9-3"></a>Where: <a id="__codelineno-9-4" name="__codelineno-9-4" href="#__codelineno-9-4"></a>- activation_score: From spreading activation (0.0-1.0) <a id="__codelineno-9-5" name="__codelineno-9-5" href="#__codelineno-9-5"></a>- decay_score: Existing temporal decay (0.0-∞) <a id="__codelineno-9-6" name="__codelineno-9-6" href="#__codelineno-9-6"></a>- review_priority: Danger zone urgency (0.0-1.0) </code></pre></div> Implementation: <div class="highlight"><pre><code><a id="__codelineno-10-1" name="__codelineno-10-1" href="#__codelineno-10-1"></a># src/cortexgraph/activation/hybrid_scoring.py <a id="__codelineno-10-2" name="__codelineno-10-2" href="#__codelineno-10-2"></a> <a id="__codelineno-10-3" name="__codelineno-10-3" href="#__codelineno-10-3"></a>from typing import List, Dict <a id="__codelineno-10-4" name="__codelineno-10-4" href="#__codelineno-10-4"></a>from ..core.decay import calculate_score <a id="__codelineno-10-5" name="__codelineno-10-5" href="#__codelineno-10-5"></a>from ..core.review import calculate_review_priority <a id="__codelineno-10-6" name="__codelineno-10-6" href="#__codelineno-10-6"></a> <a id="__codelineno-10-7" name="__codelineno-10-7" href="#__codelineno-10-7"></a>class HybridScorer: <a id="__codelineno-10-8" name="__codelineno-10-8" href="#__codelineno-10-8"></a> def __init__(self, config): <a id="__codelineno-10-9" name="__codelineno-10-9" href="#__codelineno-10-9"></a> self.activation_weight = config.ACTIVATION_WEIGHT # 0.4 default <a id="__codelineno-10-10" name="__codelineno-10-10" href="#__codelineno-10-10"></a> self.decay_weight = config.DECAY_WEIGHT # 0.4 default <a id="__codelineno-10-11" name="__codelineno-10-11" href="#__codelineno-10-11"></a> self.review_weight = config.REVIEW_WEIGHT # 0.2 default <a id="__codelineno-10-12" name="__codelineno-10-12" href="#__codelineno-10-12"></a> <a id="__codelineno-10-13" name="__codelineno-10-13" href="#__codelineno-10-13"></a> def score(self, memory, activation_score: float) -> float: <a id="__codelineno-10-14" name="__codelineno-10-14" href="#__codelineno-10-14"></a> """ <a id="__codelineno-10-15" name="__codelineno-10-15" href="#__codelineno-10-15"></a> Calculate hybrid score combining activation, decay, and review. <a id="__codelineno-10-16" name="__codelineno-10-16" href="#__codelineno-10-16"></a> <a id="__codelineno-10-17" name="__codelineno-10-17" href="#__codelineno-10-17"></a> Args: <a id="__codelineno-10-18" name="__codelineno-10-18" href="#__codelineno-10-18"></a> memory: Memory object <a id="__codelineno-10-19" name="__codelineno-10-19" href="#__codelineno-10-19"></a> activation_score: Score from spreading activation <a id="__codelineno-10-20" name="__codelineno-10-20" href="#__codelineno-10-20"></a> <a id="__codelineno-10-21" name="__codelineno-10-21" href="#__codelineno-10-21"></a> Returns: <a id="__codelineno-10-22" name="__codelineno-10-22" href="#__codelineno-10-22"></a> Combined score (0.0-∞) <a id="__codelineno-10-23" name="__codelineno-10-23" href="#__codelineno-10-23"></a> """ <a id="__codelineno-10-24" name="__codelineno-10-24" href="#__codelineno-10-24"></a> # Existing temporal decay score <a id="__codelineno-10-25" name="__codelineno-10-25" href="#__codelineno-10-25"></a> decay_score = calculate_score(memory) <a id="__codelineno-10-26" name="__codelineno-10-26" href="#__codelineno-10-26"></a> <a id="__codelineno-10-27" name="__codelineno-10-27" href="#__codelineno-10-27"></a> # Existing review priority <a id="__codelineno-10-28" name="__codelineno-10-28" href="#__codelineno-10-28"></a> review_priority = calculate_review_priority(memory) <a id="__codelineno-10-29" name="__codelineno-10-29" href="#__codelineno-10-29"></a> <a id="__codelineno-10-30" name="__codelineno-10-30" href="#__codelineno-10-30"></a> # Weighted combination <a id="__codelineno-10-31" name="__codelineno-10-31" href="#__codelineno-10-31"></a> combined = ( <a id="__codelineno-10-32" name="__codelineno-10-32" href="#__codelineno-10-32"></a> self.activation_weight * activation_score + <a id="__codelineno-10-33" name="__codelineno-10-33" href="#__codelineno-10-33"></a> self.decay_weight * decay_score + <a id="__codelineno-10-34" name="__codelineno-10-34" href="#__codelineno-10-34"></a> self.review_weight * review_priority <a id="__codelineno-10-35" name="__codelineno-10-35" href="#__codelineno-10-35"></a> ) <a id="__codelineno-10-36" name="__codelineno-10-36" href="#__codelineno-10-36"></a> <a id="__codelineno-10-37" name="__codelineno-10-37" href="#__codelineno-10-37"></a> return combined </code></pre></div> <hr /> <h4 id="component-4-natural-activation-api">Component 4: Natural Activation API<a class="headerlink" href="#component-4-natural-activation-api" title="Permanent link">¶</a></h4> Purpose: New MCP tool for conversational activation (complements existing <code>search_memory</code>) Tool Signature: <div class="highlight"><pre><code><a id="__codelineno-11-1" name="__codelineno-11-1" href="#__codelineno-11-1"></a>@mcp.tool() <a id="__codelineno-11-2" name="__codelineno-11-2" href="#__codelineno-11-2"></a>async def activate_memories( <a id="__codelineno-11-3" name="__codelineno-11-3" href="#__codelineno-11-3"></a> message: str, <a id="__codelineno-11-4" name="__codelineno-11-4" href="#__codelineno-11-4"></a> max_results: int = 5, <a id="__codelineno-11-5" name="__codelineno-11-5" href="#__codelineno-11-5"></a> include_review: bool = True <a id="__codelineno-11-6" name="__codelineno-11-6" href="#__codelineno-11-6"></a>) -> Dict: <a id="__codelineno-11-7" name="__codelineno-11-7" href="#__codelineno-11-7"></a> """ <a id="__codelineno-11-8" name="__codelineno-11-8" href="#__codelineno-11-8"></a> Automatically activate relevant memories based on conversational message. <a id="__codelineno-11-9" name="__codelineno-11-9" href="#__codelineno-11-9"></a> <a id="__codelineno-11-10" name="__codelineno-11-10" href="#__codelineno-11-10"></a> This tool uses spreading activation from entities in the message to <a id="__codelineno-11-11" name="__codelineno-11-11" href="#__codelineno-11-11"></a> surface contextually relevant memories without explicit search queries. <a id="__codelineno-11-12" name="__codelineno-11-12" href="#__codelineno-11-12"></a> <a id="__codelineno-11-13" name="__codelineno-11-13" href="#__codelineno-11-13"></a> Args: <a id="__codelineno-11-14" name="__codelineno-11-14" href="#__codelineno-11-14"></a> message: User's conversational message <a id="__codelineno-11-15" name="__codelineno-11-15" href="#__codelineno-11-15"></a> max_results: Maximum memories to return (default: 5) <a id="__codelineno-11-16" name="__codelineno-11-16" href="#__codelineno-11-16"></a> include_review: Blend in review candidates (default: True) <a id="__codelineno-11-17" name="__codelineno-11-17" href="#__codelineno-11-17"></a> <a id="__codelineno-11-18" name="__codelineno-11-18" href="#__codelineno-11-18"></a> Returns: <a id="__codelineno-11-19" name="__codelineno-11-19" href="#__codelineno-11-19"></a> { <a id="__codelineno-11-20" name="__codelineno-11-20" href="#__codelineno-11-20"></a> "activated_memories": [ <a id="__codelineno-11-21" name="__codelineno-11-21" href="#__codelineno-11-21"></a> { <a id="__codelineno-11-22" name="__codelineno-11-22" href="#__codelineno-11-22"></a> "id": "mem-123", <a id="__codelineno-11-23" name="__codelineno-11-23" href="#__codelineno-11-23"></a> "content": "...", <a id="__codelineno-11-24" name="__codelineno-11-24" href="#__codelineno-11-24"></a> "activation_score": 0.85, <a id="__codelineno-11-25" name="__codelineno-11-25" href="#__codelineno-11-25"></a> "decay_score": 0.62, <a id="__codelineno-11-26" name="__codelineno-11-26" href="#__codelineno-11-26"></a> "hybrid_score": 0.73, <a id="__codelineno-11-27" name="__codelineno-11-27" href="#__codelineno-11-27"></a> "source": "spreading_activation" | "review_candidate", <a id="__codelineno-11-28" name="__codelineno-11-28" href="#__codelineno-11-28"></a> "activation_path": ["entity:typescript", "relation:prefers", "mem-123"] <a id="__codelineno-11-29" name="__codelineno-11-29" href="#__codelineno-11-29"></a> }, <a id="__codelineno-11-30" name="__codelineno-11-30" href="#__codelineno-11-30"></a> ... <a id="__codelineno-11-31" name="__codelineno-11-31" href="#__codelineno-11-31"></a> ], <a id="__codelineno-11-32" name="__codelineno-11-32" href="#__codelineno-11-32"></a> "trigger_info": { <a id="__codelineno-11-33" name="__codelineno-11-33" href="#__codelineno-11-33"></a> "should_activate": True, <a id="__codelineno-11-34" name="__codelineno-11-34" href="#__codelineno-11-34"></a> "activation_type": "question", <a id="__codelineno-11-35" name="__codelineno-11-35" href="#__codelineno-11-35"></a> "source_entities": ["typescript", "backend"], <a id="__codelineno-11-36" name="__codelineno-11-36" href="#__codelineno-11-36"></a> "intent": "recall" <a id="__codelineno-11-37" name="__codelineno-11-37" href="#__codelineno-11-37"></a> }, <a id="__codelineno-11-38" name="__codelineno-11-38" href="#__codelineno-11-38"></a> "stats": { <a id="__codelineno-11-39" name="__codelineno-11-39" href="#__codelineno-11-39"></a> "total_activated": 12, <a id="__codelineno-11-40" name="__codelineno-11-40" href="#__codelineno-11-40"></a> "returned_count": 5, <a id="__codelineno-11-41" name="__codelineno-11-41" href="#__codelineno-11-41"></a> "activation_hops": 3 <a id="__codelineno-11-42" name="__codelineno-11-42" href="#__codelineno-11-42"></a> } <a id="__codelineno-11-43" name="__codelineno-11-43" href="#__codelineno-11-43"></a> } <a id="__codelineno-11-44" name="__codelineno-11-44" href="#__codelineno-11-44"></a> """ <a id="__codelineno-11-45" name="__codelineno-11-45" href="#__codelineno-11-45"></a> # Implementation in tools/activate.py <a id="__codelineno-11-46" name="__codelineno-11-46" href="#__codelineno-11-46"></a> ... </code></pre></div> Usage Example: <div class="highlight"><pre><code><a id="__codelineno-12-1" name="__codelineno-12-1" href="#__codelineno-12-1"></a># User: "What did I decide about TypeScript for backend projects?" <a id="__codelineno-12-2" name="__codelineno-12-2" href="#__codelineno-12-2"></a> <a id="__codelineno-12-3" name="__codelineno-12-3" href="#__codelineno-12-3"></a>result = activate_memories( <a id="__codelineno-12-4" name="__codelineno-12-4" href="#__codelineno-12-4"></a> message="What did I decide about TypeScript for backend projects?", <a id="__codelineno-12-5" name="__codelineno-12-5" href="#__codelineno-12-5"></a> max_results=5 <a id="__codelineno-12-6" name="__codelineno-12-6" href="#__codelineno-12-6"></a>) <a id="__codelineno-12-7" name="__codelineno-12-7" href="#__codelineno-12-7"></a> <a id="__codelineno-12-8" name="__codelineno-12-8" href="#__codelineno-12-8"></a># Returns: <a id="__codelineno-12-9" name="__codelineno-12-9" href="#__codelineno-12-9"></a># - Memory about TypeScript preference (direct entity match) <a id="__codelineno-12-10" name="__codelineno-12-10" href="#__codelineno-12-10"></a># - Memory about backend architecture choice (1-hop relation) <a id="__codelineno-12-11" name="__codelineno-12-11" href="#__codelineno-12-11"></a># - Memory about related project X (2-hop relation via shared entity) <a id="__codelineno-12-12" name="__codelineno-12-12" href="#__codelineno-12-12"></a># - Review candidate about JWT authentication (danger zone, tag overlap) </code></pre></div> <hr /> <h2 id="implementation-phases">Implementation Phases<a class="headerlink" href="#implementation-phases" title="Permanent link">¶</a></h2> <h3 id="phase-1-core-spreading-activation-3-weeks">Phase 1: Core Spreading Activation (3 weeks)<a class="headerlink" href="#phase-1-core-spreading-activation-3-weeks" title="Permanent link">¶</a></h3> Goal: Implement basic spreading activation on existing knowledge graph Deliverables: - ✅ <code>src/cortexgraph/activation/spreading.py</code> - Core activation algorithm - ✅ <code>src/cortexgraph/activation/trigger_detector.py</code> - Entity extraction + intent detection - ✅ <code>src/cortexgraph/activation/hybrid_scoring.py</code> - Combine activation + decay + review - ✅ <code>tests/activation/test_spreading.py</code> - Unit tests (90%+ coverage) - ✅ Configuration options in <code>config.py</code> Success Criteria: - ✅ Activate memories through 1-3 hops in knowledge graph - ✅ Combine activation scores with temporal decay correctly - ✅ Entity extraction works on 80%+ of test messages Dependencies: - Existing knowledge graph (entities + relations) ✅ - spaCy for NER (<code>en_core_web_sm</code>) - new <hr /> <h3 id="phase-2-mcp-tool-integration-2-weeks">Phase 2: MCP Tool Integration (2 weeks)<a class="headerlink" href="#phase-2-mcp-tool-integration-2-weeks" title="Permanent link">¶</a></h3> Goal: Expose spreading activation via MCP tool Deliverables: - ✅ <code>src/cortexgraph/tools/activate.py</code> - New <code>activate_memories</code> tool - ✅ Integration with MCP server (<code>server.py</code>) - ✅ Documentation in <code>docs/api.md</code> - ✅ Integration tests (end-to-end flow) Success Criteria: - ✅ LLM can call <code>activate_memories</code> from conversation - ✅ Returns relevant memories without explicit search query - ✅ Activation explanations (path tracing) included in response <hr /> <h3 id="phase-3-advanced-features-3-weeks">Phase 3: Advanced Features (3 weeks)<a class="headerlink" href="#phase-3-advanced-features-3-weeks" title="Permanent link">¶</a></h3> Goal: Category inference, automatic relation creation, multi-dimensional search Component 3.1: Category Inference <div class="highlight"><pre><code><a id="__codelineno-13-1" name="__codelineno-13-1" href="#__codelineno-13-1"></a># src/cortexgraph/activation/categorizer.py <a id="__codelineno-13-2" name="__codelineno-13-2" href="#__codelineno-13-2"></a> <a id="__codelineno-13-3" name="__codelineno-13-3" href="#__codelineno-13-3"></a>from transformers import pipeline <a id="__codelineno-13-4" name="__codelineno-13-4" href="#__codelineno-13-4"></a> <a id="__codelineno-13-5" name="__codelineno-13-5" href="#__codelineno-13-5"></a>class CategoryInference: <a id="__codelineno-13-6" name="__codelineno-13-6" href="#__codelineno-13-6"></a> def __init__(self): <a id="__codelineno-13-7" name="__codelineno-13-7" href="#__codelineno-13-7"></a> # Zero-shot classification for predefined categories <a id="__codelineno-13-8" name="__codelineno-13-8" href="#__codelineno-13-8"></a> self.classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli") <a id="__codelineno-13-9" name="__codelineno-13-9" href="#__codelineno-13-9"></a> <a id="__codelineno-13-10" name="__codelineno-13-10" href="#__codelineno-13-10"></a> self.categories = [ <a id="__codelineno-13-11" name="__codelineno-13-11" href="#__codelineno-13-11"></a> "programming", "project-management", "preferences", <a id="__codelineno-13-12" name="__codelineno-13-12" href="#__codelineno-13-12"></a> "decisions", "facts", "relationships", "goals" <a id="__codelineno-13-13" name="__codelineno-13-13" href="#__codelineno-13-13"></a> ] <a id="__codelineno-13-14" name="__codelineno-13-14" href="#__codelineno-13-14"></a> <a id="__codelineno-13-15" name="__codelineno-13-15" href="#__codelineno-13-15"></a> def infer_categories(self, content: str) -> List[str]: <a id="__codelineno-13-16" name="__codelineno-13-16" href="#__codelineno-13-16"></a> """ <a id="__codelineno-13-17" name="__codelineno-13-17" href="#__codelineno-13-17"></a> Classify memory content into predefined categories. <a id="__codelineno-13-18" name="__codelineno-13-18" href="#__codelineno-13-18"></a> <a id="__codelineno-13-19" name="__codelineno-13-19" href="#__codelineno-13-19"></a> Returns categories with confidence > 0.5 <a id="__codelineno-13-20" name="__codelineno-13-20" href="#__codelineno-13-20"></a> """ <a id="__codelineno-13-21" name="__codelineno-13-21" href="#__codelineno-13-21"></a> result = self.classifier(content, self.categories, multi_label=True) <a id="__codelineno-13-22" name="__codelineno-13-22" href="#__codelineno-13-22"></a> return [label for label, score in zip(result["labels"], result["scores"]) if score > 0.5] </code></pre></div> Component 3.2: Automatic Relation Creation <div class="highlight"><pre><code><a id="__codelineno-14-1" name="__codelineno-14-1" href="#__codelineno-14-1"></a># src/cortexgraph/activation/auto_relations.py <a id="__codelineno-14-2" name="__codelineno-14-2" href="#__codelineno-14-2"></a> <a id="__codelineno-14-3" name="__codelineno-14-3" href="#__codelineno-14-3"></a>class AutoRelationCreator: <a id="__codelineno-14-4" name="__codelineno-14-4" href="#__codelineno-14-4"></a> def __init__(self, storage, similarity_threshold=0.85): <a id="__codelineno-14-5" name="__codelineno-14-5" href="#__codelineno-14-5"></a> self.storage = storage <a id="__codelineno-14-6" name="__codelineno-14-6" href="#__codelineno-14-6"></a> self.threshold = similarity_threshold <a id="__codelineno-14-7" name="__codelineno-14-7" href="#__codelineno-14-7"></a> <a id="__codelineno-14-8" name="__codelineno-14-8" href="#__codelineno-14-8"></a> async def create_relations_for_new_memory(self, memory_id: str): <a id="__codelineno-14-9" name="__codelineno-14-9" href="#__codelineno-14-9"></a> """ <a id="__codelineno-14-10" name="__codelineno-14-10" href="#__codelineno-14-10"></a> Automatically create relations to similar/related existing memories. <a id="__codelineno-14-11" name="__codelineno-14-11" href="#__codelineno-14-11"></a> <a id="__codelineno-14-12" name="__codelineno-14-12" href="#__codelineno-14-12"></a> Uses: <a id="__codelineno-14-13" name="__codelineno-14-13" href="#__codelineno-14-13"></a> - Entity overlap (shared entities → "related_to") <a id="__codelineno-14-14" name="__codelineno-14-14" href="#__codelineno-14-14"></a> - Semantic similarity (embeddings → "similar_to") <a id="__codelineno-14-15" name="__codelineno-14-15" href="#__codelineno-14-15"></a> - Temporal proximity (created within 24h → "follows_from") <a id="__codelineno-14-16" name="__codelineno-14-16" href="#__codelineno-14-16"></a> """ <a id="__codelineno-14-17" name="__codelineno-14-17" href="#__codelineno-14-17"></a> new_memory = self.storage.get_memory(memory_id) <a id="__codelineno-14-18" name="__codelineno-14-18" href="#__codelineno-14-18"></a> candidates = self.storage.get_all_memories(status="active") <a id="__codelineno-14-19" name="__codelineno-14-19" href="#__codelineno-14-19"></a> <a id="__codelineno-14-20" name="__codelineno-14-20" href="#__codelineno-14-20"></a> for candidate in candidates: <a id="__codelineno-14-21" name="__codelineno-14-21" href="#__codelineno-14-21"></a> if candidate.id == memory_id: <a id="__codelineno-14-22" name="__codelineno-14-22" href="#__codelineno-14-22"></a> continue <a id="__codelineno-14-23" name="__codelineno-14-23" href="#__codelineno-14-23"></a> <a id="__codelineno-14-24" name="__codelineno-14-24" href="#__codelineno-14-24"></a> # Check entity overlap <a id="__codelineno-14-25" name="__codelineno-14-25" href="#__codelineno-14-25"></a> shared_entities = set(new_memory.entities) & set(candidate.entities) <a id="__codelineno-14-26" name="__codelineno-14-26" href="#__codelineno-14-26"></a> if len(shared_entities) >= 2: <a id="__codelineno-14-27" name="__codelineno-14-27" href="#__codelineno-14-27"></a> await self.storage.create_relation( <a id="__codelineno-14-28" name="__codelineno-14-28" href="#__codelineno-14-28"></a> from_id=memory_id, <a id="__codelineno-14-29" name="__codelineno-14-29" href="#__codelineno-14-29"></a> to_id=candidate.id, <a id="__codelineno-14-30" name="__codelineno-14-30" href="#__codelineno-14-30"></a> relation_type="related_to", <a id="__codelineno-14-31" name="__codelineno-14-31" href="#__codelineno-14-31"></a> strength=0.7, <a id="__codelineno-14-32" name="__codelineno-14-32" href="#__codelineno-14-32"></a> metadata={"shared_entities": list(shared_entities), "auto_created": True} <a id="__codelineno-14-33" name="__codelineno-14-33" href="#__codelineno-14-33"></a> ) <a id="__codelineno-14-34" name="__codelineno-14-34" href="#__codelineno-14-34"></a> <a id="__codelineno-14-35" name="__codelineno-14-35" href="#__codelineno-14-35"></a> # Check semantic similarity (if embeddings enabled) <a id="__codelineno-14-36" name="__codelineno-14-36" href="#__codelineno-14-36"></a> if new_memory.embedding and candidate.embedding: <a id="__codelineno-14-37" name="__codelineno-14-37" href="#__codelineno-14-37"></a> similarity = cosine_similarity(new_memory.embedding, candidate.embedding) <a id="__codelineno-14-38" name="__codelineno-14-38" href="#__codelineno-14-38"></a> if similarity > self.threshold: <a id="__codelineno-14-39" name="__codelineno-14-39" href="#__codelineno-14-39"></a> await self.storage.create_relation( <a id="__codelineno-14-40" name="__codelineno-14-40" href="#__codelineno-14-40"></a> from_id=memory_id, <a id="__codelineno-14-41" name="__codelineno-14-41" href="#__codelineno-14-41"></a> to_id=candidate.id, <a id="__codelineno-14-42" name="__codelineno-14-42" href="#__codelineno-14-42"></a> relation_type="similar_to", <a id="__codelineno-14-43" name="__codelineno-14-43" href="#__codelineno-14-43"></a> strength=similarity, <a id="__codelineno-14-44" name="__codelineno-14-44" href="#__codelineno-14-44"></a> metadata={"similarity_score": similarity, "auto_created": True} <a id="__codelineno-14-45" name="__codelineno-14-45" href="#__codelineno-14-45"></a> ) </code></pre></div> Component 3.3: Multi-Dimensional Search Extend existing <code>search_memory</code> with IMDMR-inspired dimensions: <div class="highlight"><pre><code><a id="__codelineno-15-1" name="__codelineno-15-1" href="#__codelineno-15-1"></a># Enhance search_memory to support multi-dimensional ranking <a id="__codelineno-15-2" name="__codelineno-15-2" href="#__codelineno-15-2"></a> <a id="__codelineno-15-3" name="__codelineno-15-3" href="#__codelineno-15-3"></a>def search_memory_multidim( <a id="__codelineno-15-4" name="__codelineno-15-4" href="#__codelineno-15-4"></a> query: str, <a id="__codelineno-15-5" name="__codelineno-15-5" href="#__codelineno-15-5"></a> tags: List[str] = None, <a id="__codelineno-15-6" name="__codelineno-15-6" href="#__codelineno-15-6"></a> entities: List[str] = None, <a id="__codelineno-15-7" name="__codelineno-15-7" href="#__codelineno-15-7"></a> categories: List[str] = None, <a id="__codelineno-15-8" name="__codelineno-15-8" href="#__codelineno-15-8"></a> intent: str = None, <a id="__codelineno-15-9" name="__codelineno-15-9" href="#__codelineno-15-9"></a> top_k: int = 10 <a id="__codelineno-15-10" name="__codelineno-15-10" href="#__codelineno-15-10"></a>) -> List[Memory]: <a id="__codelineno-15-11" name="__codelineno-15-11" href="#__codelineno-15-11"></a> """ <a id="__codelineno-15-12" name="__codelineno-15-12" href="#__codelineno-15-12"></a> Multi-dimensional memory search combining: <a id="__codelineno-15-13" name="__codelineno-15-13" href="#__codelineno-15-13"></a> - Semantic: embedding similarity <a id="__codelineno-15-14" name="__codelineno-15-14" href="#__codelineno-15-14"></a> - Entity: entity overlap <a id="__codelineno-15-15" name="__codelineno-15-15" href="#__codelineno-15-15"></a> - Category: category match <a id="__codelineno-15-16" name="__codelineno-15-16" href="#__codelineno-15-16"></a> - Intent: intent alignment <a id="__codelineno-15-17" name="__codelineno-15-17" href="#__codelineno-15-17"></a> - Temporal: decay score <a id="__codelineno-15-18" name="__codelineno-15-18" href="#__codelineno-15-18"></a> - Context: tag overlap <a id="__codelineno-15-19" name="__codelineno-15-19" href="#__codelineno-15-19"></a> """ <a id="__codelineno-15-20" name="__codelineno-15-20" href="#__codelineno-15-20"></a> # Score each dimension separately <a id="__codelineno-15-21" name="__codelineno-15-21" href="#__codelineno-15-21"></a> semantic_scores = _score_semantic(query, candidates) <a id="__codelineno-15-22" name="__codelineno-15-22" href="#__codelineno-15-22"></a> entity_scores = _score_entity_overlap(entities, candidates) <a id="__codelineno-15-23" name="__codelineno-15-23" href="#__codelineno-15-23"></a> category_scores = _score_category_match(categories, candidates) <a id="__codelineno-15-24" name="__codelineno-15-24" href="#__codelineno-15-24"></a> temporal_scores = _score_temporal_decay(candidates) <a id="__codelineno-15-25" name="__codelineno-15-25" href="#__codelineno-15-25"></a> context_scores = _score_tag_overlap(tags, candidates) <a id="__codelineno-15-26" name="__codelineno-15-26" href="#__codelineno-15-26"></a> <a id="__codelineno-15-27" name="__codelineno-15-27" href="#__codelineno-15-27"></a> # Weighted combination <a id="__codelineno-15-28" name="__codelineno-15-28" href="#__codelineno-15-28"></a> final_scores = ( <a id="__codelineno-15-29" name="__codelineno-15-29" href="#__codelineno-15-29"></a> 0.3 * semantic_scores + <a id="__codelineno-15-30" name="__codelineno-15-30" href="#__codelineno-15-30"></a> 0.2 * entity_scores + <a id="__codelineno-15-31" name="__codelineno-15-31" href="#__codelineno-15-31"></a> 0.15 * category_scores + <a id="__codelineno-15-32" name="__codelineno-15-32" href="#__codelineno-15-32"></a> 0.2 * temporal_scores + <a id="__codelineno-15-33" name="__codelineno-15-33" href="#__codelineno-15-33"></a> 0.15 * context_scores <a id="__codelineno-15-34" name="__codelineno-15-34" href="#__codelineno-15-34"></a> ) <a id="__codelineno-15-35" name="__codelineno-15-35" href="#__codelineno-15-35"></a> <a id="__codelineno-15-36" name="__codelineno-15-36" href="#__codelineno-15-36"></a> # Rank and return top-k <a id="__codelineno-15-37" name="__codelineno-15-37" href="#__codelineno-15-37"></a> return sorted(candidates, key=lambda m: final_scores[m.id], reverse=True)[:top_k] </code></pre></div> Deliverables: - ✅ Category inference (zero-shot classification) - ✅ Automatic relation creation on <code>save_memory</code> - ✅ Multi-dimensional search enhancement - ✅ Tests for each component Success Criteria: - ✅ Categories automatically inferred with 70%+ accuracy - ✅ Auto-relations reduce manual linking effort by 60%+ - ✅ Multi-dimensional search outperforms single-dimension by 20%+ <hr /> <h3 id="phase-4-production-tuning-2-weeks">Phase 4: Production Tuning (2 weeks)<a class="headerlink" href="#phase-4-production-tuning-2-weeks" title="Permanent link">¶</a></h3> Goal: Performance optimization, configuration tuning, user testing Optimization Targets: - Activation latency < 100ms (in-memory graph traversal) - Category inference < 50ms (lightweight model) - Auto-relation creation async (doesn't block save_memory) Configuration Tuning: <div class="highlight"><pre><code><a id="__codelineno-16-1" name="__codelineno-16-1" href="#__codelineno-16-1"></a># Spreading Activation <a id="__codelineno-16-2" name="__codelineno-16-2" href="#__codelineno-16-2"></a>ACTIVATION_DECAY_RATE=0.7 # Test 0.6, 0.7, 0.8 <a id="__codelineno-16-3" name="__codelineno-16-3" href="#__codelineno-16-3"></a>ACTIVATION_THRESHOLD=0.15 # Test 0.10, 0.15, 0.20 <a id="__codelineno-16-4" name="__codelineno-16-4" href="#__codelineno-16-4"></a>MAX_ACTIVATION_HOPS=3 # Test 2, 3, 4 <a id="__codelineno-16-5" name="__codelineno-16-5" href="#__codelineno-16-5"></a> <a id="__codelineno-16-6" name="__codelineno-16-6" href="#__codelineno-16-6"></a># Hybrid Scoring Weights <a id="__codelineno-16-7" name="__codelineno-16-7" href="#__codelineno-16-7"></a>ACTIVATION_WEIGHT=0.4 # Test 0.3-0.5 <a id="__codelineno-16-8" name="__codelineno-16-8" href="#__codelineno-16-8"></a>DECAY_WEIGHT=0.4 # Test 0.3-0.5 <a id="__codelineno-16-9" name="__codelineno-16-9" href="#__codelineno-16-9"></a>REVIEW_WEIGHT=0.2 # Test 0.1-0.3 <a id="__codelineno-16-10" name="__codelineno-16-10" href="#__codelineno-16-10"></a> <a id="__codelineno-16-11" name="__codelineno-16-11" href="#__codelineno-16-11"></a># Auto Relations <a id="__codelineno-16-12" name="__codelineno-16-12" href="#__codelineno-16-12"></a>AUTO_RELATION_ENABLED=true <a id="__codelineno-16-13" name="__codelineno-16-13" href="#__codelineno-16-13"></a>AUTO_RELATION_MIN_ENTITY_OVERLAP=2 <a id="__codelineno-16-14" name="__codelineno-16-14" href="#__codelineno-16-14"></a>AUTO_RELATION_SIMILARITY_THRESHOLD=0.85 </code></pre></div> User Testing: - A/B test: Activation ON vs. OFF - Metrics: Conversation quality, memory recall accuracy, user satisfaction - Target: 3-4x improvement in relevant memory retrieval Deliverables: - ✅ Performance benchmarks - ✅ Configuration recommendations - ✅ User testing report - ✅ Documentation updates <hr /> <h2 id="integration-points">Integration Points<a class="headerlink" href="#integration-points" title="Permanent link">¶</a></h2> <h3 id="1-mcp-server-entry-point">1. MCP Server Entry Point<a class="headerlink" href="#1-mcp-server-entry-point" title="Permanent link">¶</a></h3> File: <code>src/cortexgraph/server.py</code> <div class="highlight"><pre><code><a id="__codelineno-17-1" name="__codelineno-17-1" href="#__codelineno-17-1"></a>from .activation import ActivationTrigger, SpreadingActivation, HybridScorer <a id="__codelineno-17-2" name="__codelineno-17-2" href="#__codelineno-17-2"></a> <a id="__codelineno-17-3" name="__codelineno-17-3" href="#__codelineno-17-3"></a># Initialize activation components (lazy loading) <a id="__codelineno-17-4" name="__codelineno-17-4" href="#__codelineno-17-4"></a>_activation_components = None <a id="__codelineno-17-5" name="__codelineno-17-5" href="#__codelineno-17-5"></a> <a id="__codelineno-17-6" name="__codelineno-17-6" href="#__codelineno-17-6"></a>def get_activation_components(): <a id="__codelineno-17-7" name="__codelineno-17-7" href="#__codelineno-17-7"></a> global _activation_components <a id="__codelineno-17-8" name="__codelineno-17-8" href="#__codelineno-17-8"></a> if _activation_components is None and config.ENABLE_ACTIVATION: <a id="__codelineno-17-9" name="__codelineno-17-9" href="#__codelineno-17-9"></a> _activation_components = { <a id="__codelineno-17-10" name="__codelineno-17-10" href="#__codelineno-17-10"></a> "trigger": ActivationTrigger(), <a id="__codelineno-17-11" name="__codelineno-17-11" href="#__codelineno-17-11"></a> "spreader": SpreadingActivation(storage, config), <a id="__codelineno-17-12" name="__codelineno-17-12" href="#__codelineno-17-12"></a> "scorer": HybridScorer(config), <a id="__codelineno-17-13" name="__codelineno-17-13" href="#__codelineno-17-13"></a> } <a id="__codelineno-17-14" name="__codelineno-17-14" href="#__codelineno-17-14"></a> return _activation_components <a id="__codelineno-17-15" name="__codelineno-17-15" href="#__codelineno-17-15"></a> <a id="__codelineno-17-16" name="__codelineno-17-16" href="#__codelineno-17-16"></a>@mcp.tool() <a id="__codelineno-17-17" name="__codelineno-17-17" href="#__codelineno-17-17"></a>async def activate_memories(message: str, max_results: int = 5, include_review: bool = True): <a id="__codelineno-17-18" name="__codelineno-17-18" href="#__codelineno-17-18"></a> """Natural language memory activation.""" <a id="__codelineno-17-19" name="__codelineno-17-19" href="#__codelineno-17-19"></a> # Implementation calls components above <a id="__codelineno-17-20" name="__codelineno-17-20" href="#__codelineno-17-20"></a> ... </code></pre></div> <h3 id="2-integration-with-existing-tools">2. Integration with Existing Tools<a class="headerlink" href="#2-integration-with-existing-tools" title="Permanent link">¶</a></h3> Relationship to <code>search_memory</code>: - <code>activate_memories</code>: Automatic, conversational, graph-based - <code>search_memory</code>: Explicit, query-driven, text/tag-based - Both can coexist and complement each other Enhancement to <code>save_memory</code>: <div class="highlight"><pre><code><a id="__codelineno-18-1" name="__codelineno-18-1" href="#__codelineno-18-1"></a>@mcp.tool() <a id="__codelineno-18-2" name="__codelineno-18-2" href="#__codelineno-18-2"></a>async def save_memory(content, tags, entities, ...): <a id="__codelineno-18-3" name="__codelineno-18-3" href="#__codelineno-18-3"></a> # Existing save logic <a id="__codelineno-18-4" name="__codelineno-18-4" href="#__codelineno-18-4"></a> memory_id = storage.save(...) <a id="__codelineno-18-5" name="__codelineno-18-5" href="#__codelineno-18-5"></a> <a id="__codelineno-18-6" name="__codelineno-18-6" href="#__codelineno-18-6"></a> # NEW: Automatic relation creation <a id="__codelineno-18-7" name="__codelineno-18-7" href="#__codelineno-18-7"></a> if config.AUTO_RELATION_ENABLED: <a id="__codelineno-18-8" name="__codelineno-18-8" href="#__codelineno-18-8"></a> await auto_relation_creator.create_relations_for_new_memory(memory_id) <a id="__codelineno-18-9" name="__codelineno-18-9" href="#__codelineno-18-9"></a> <a id="__codelineno-18-10" name="__codelineno-18-10" href="#__codelineno-18-10"></a> # NEW: Category inference <a id="__codelineno-18-11" name="__codelineno-18-11" href="#__codelineno-18-11"></a> if config.ENABLE_CATEGORY_INFERENCE: <a id="__codelineno-18-12" name="__codelineno-18-12" href="#__codelineno-18-12"></a> categories = categorizer.infer_categories(content) <a id="__codelineno-18-13" name="__codelineno-18-13" href="#__codelineno-18-13"></a> storage.update_categories(memory_id, categories) <a id="__codelineno-18-14" name="__codelineno-18-14" href="#__codelineno-18-14"></a> <a id="__codelineno-18-15" name="__codelineno-18-15" href="#__codelineno-18-15"></a> return memory_id </code></pre></div> <h3 id="3-conversational-activation-integration">3. Conversational Activation Integration<a class="headerlink" href="#3-conversational-activation-integration" title="Permanent link">¶</a></h3> This feature complements the approved conversational-activation-plan.md: conversational-activation-plan.md: Preprocessing layer for detecting when to save - Intent classification (SAVE_PREFERENCE, SAVE_DECISION, etc.) - Entity extraction for populating <code>entities</code> field - Tag suggestion - Importance scoring natural-language-activation (this plan): Graph traversal for retrieving related memories - Spreading activation from entities - Multi-dimensional search - Automatic relation creation Together: Complete conversational memory system <div class="highlight"><pre><code><a id="__codelineno-19-1" name="__codelineno-19-1" href="#__codelineno-19-1"></a>User Message <a id="__codelineno-19-2" name="__codelineno-19-2" href="#__codelineno-19-2"></a> ↓ <a id="__codelineno-19-3" name="__codelineno-19-3" href="#__codelineno-19-3"></a>Preprocessing (conversational-activation-plan) <a id="__codelineno-19-4" name="__codelineno-19-4" href="#__codelineno-19-4"></a> ├─ Intent: SAVE_PREFERENCE → save_memory <a id="__codelineno-19-5" name="__codelineno-19-5" href="#__codelineno-19-5"></a> └─ Intent: RECALL_INFO → activate_memories (this plan) <a id="__codelineno-19-6" name="__codelineno-19-6" href="#__codelineno-19-6"></a> ↓ <a id="__codelineno-19-7" name="__codelineno-19-7" href="#__codelineno-19-7"></a>Memory Operations </code></pre></div> <hr /> <h2 id="success-metrics">Success Metrics<a class="headerlink" href="#success-metrics" title="Permanent link">¶</a></h2> <h3 id="quantitative-metrics">Quantitative Metrics<a class="headerlink" href="#quantitative-metrics" title="Permanent link">¶</a></h3> 1. Activation Quality (Primary Metric): - Baseline: Explicit search (<code>search_memory</code>) retrieves 2-3 relevant memories per query - Target: Spreading activation retrieves 4-8 relevant memories per conversation turn - Measurement: Manual annotation of relevance (human judgment) 2. Precision/Recall: - Precision: % of activated memories that are relevant - Target: 70%+ (vs. 85%+ for explicit search - acceptable tradeoff for breadth) - Recall: % of relevant memories that are activated - Target: 80%+ (vs. 60% for explicit search - improvement through graph traversal) 3. Latency: - Activation time: < 100ms (in-memory graph traversal) - Total retrieval time: < 200ms (activation + scoring + ranking) 4. Graph Density: - Auto-relations created: 60%+ reduction in manual relation effort - Average relations per memory: Increase from ~0.5 to ~2.5 <h3 id="qualitative-metrics">Qualitative Metrics<a class="headerlink" href="#qualitative-metrics" title="Permanent link">¶</a></h3> User Experience: - Survey: "Does the AI remember context naturally?" (8/10 target) - Survey: "How often does the AI miss relevant information?" (Rarely/Never target) Developer Experience: - Ease of configuration (tuning activation parameters) - Debuggability (activation path tracing) <hr /> <h2 id="configuration">Configuration<a class="headerlink" href="#configuration" title="Permanent link">¶</a></h2> <div class="highlight"><pre><code><a id="__codelineno-20-1" name="__codelineno-20-1" href="#__codelineno-20-1"></a># ============================================================================ <a id="__codelineno-20-2" name="__codelineno-20-2" href="#__codelineno-20-2"></a># Natural Language Activation Configuration <a id="__codelineno-20-3" name="__codelineno-20-3" href="#__codelineno-20-3"></a># ============================================================================ <a id="__codelineno-20-4" name="__codelineno-20-4" href="#__codelineno-20-4"></a> <a id="__codelineno-20-5" name="__codelineno-20-5" href="#__codelineno-20-5"></a># Enable/Disable Activation <a id="__codelineno-20-6" name="__codelineno-20-6" href="#__codelineno-20-6"></a>ENABLE_ACTIVATION=true <a id="__codelineno-20-7" name="__codelineno-20-7" href="#__codelineno-20-7"></a> <a id="__codelineno-20-8" name="__codelineno-20-8" href="#__codelineno-20-8"></a># Spreading Activation <a id="__codelineno-20-9" name="__codelineno-20-9" href="#__codelineno-20-9"></a>ACTIVATION_DECAY_RATE=0.7 # Activation strength decay per hop (0.0-1.0) <a id="__codelineno-20-10" name="__codelineno-20-10" href="#__codelineno-20-10"></a>ACTIVATION_THRESHOLD=0.15 # Minimum activation to consider memory <a id="__codelineno-20-11" name="__codelineno-20-11" href="#__codelineno-20-11"></a>MAX_ACTIVATION_HOPS=3 # Maximum graph traversal depth (1-5) <a id="__codelineno-20-12" name="__codelineno-20-12" href="#__codelineno-20-12"></a> <a id="__codelineno-20-13" name="__codelineno-20-13" href="#__codelineno-20-13"></a># Hybrid Scoring Weights <a id="__codelineno-20-14" name="__codelineno-20-14" href="#__codelineno-20-14"></a>ACTIVATION_WEIGHT=0.4 # Weight for activation score (0.0-1.0) <a id="__codelineno-20-15" name="__codelineno-20-15" href="#__codelineno-20-15"></a>DECAY_WEIGHT=0.4 # Weight for temporal decay (0.0-1.0) <a id="__codelineno-20-16" name="__codelineno-20-16" href="#__codelineno-20-16"></a>REVIEW_WEIGHT=0.2 # Weight for review priority (0.0-1.0) <a id="__codelineno-20-17" name="__codelineno-20-17" href="#__codelineno-20-17"></a> <a id="__codelineno-20-18" name="__codelineno-20-18" href="#__codelineno-20-18"></a># Automatic Relation Creation <a id="__codelineno-20-19" name="__codelineno-20-19" href="#__codelineno-20-19"></a>AUTO_RELATION_ENABLED=true <a id="__codelineno-20-20" name="__codelineno-20-20" href="#__codelineno-20-20"></a>AUTO_RELATION_MIN_ENTITY_OVERLAP=2 # Min shared entities for "related_to" <a id="__codelineno-20-21" name="__codelineno-20-21" href="#__codelineno-20-21"></a>AUTO_RELATION_SIMILARITY_THRESHOLD=0.85 # Min similarity for "similar_to" <a id="__codelineno-20-22" name="__codelineno-20-22" href="#__codelineno-20-22"></a> <a id="__codelineno-20-23" name="__codelineno-20-23" href="#__codelineno-20-23"></a># Category Inference <a id="__codelineno-20-24" name="__codelineno-20-24" href="#__codelineno-20-24"></a>ENABLE_CATEGORY_INFERENCE=true <a id="__codelineno-20-25" name="__codelineno-20-25" href="#__codelineno-20-25"></a>CATEGORY_MODEL=facebook/bart-large-mnli # Zero-shot classification model <a id="__codelineno-20-26" name="__codelineno-20-26" href="#__codelineno-20-26"></a>CATEGORY_CONFIDENCE_THRESHOLD=0.5 # Min confidence to assign category <a id="__codelineno-20-27" name="__codelineno-20-27" href="#__codelineno-20-27"></a> <a id="__codelineno-20-28" name="__codelineno-20-28" href="#__codelineno-20-28"></a># Multi-Dimensional Search <a id="__codelineno-20-29" name="__codelineno-20-29" href="#__codelineno-20-29"></a>ENABLE_MULTIDIM_SEARCH=true <a id="__codelineno-20-30" name="__codelineno-20-30" href="#__codelineno-20-30"></a>MULTIDIM_SEMANTIC_WEIGHT=0.3 <a id="__codelineno-20-31" name="__codelineno-20-31" href="#__codelineno-20-31"></a>MULTIDIM_ENTITY_WEIGHT=0.2 <a id="__codelineno-20-32" name="__codelineno-20-32" href="#__codelineno-20-32"></a>MULTIDIM_CATEGORY_WEIGHT=0.15 <a id="__codelineno-20-33" name="__codelineno-20-33" href="#__codelineno-20-33"></a>MULTIDIM_TEMPORAL_WEIGHT=0.2 <a id="__codelineno-20-34" name="__codelineno-20-34" href="#__codelineno-20-34"></a>MULTIDIM_CONTEXT_WEIGHT=0.15 </code></pre></div> <hr /> <h2 id="dependencies">Dependencies<a class="headerlink" href="#dependencies" title="Permanent link">¶</a></h2> <h3 id="python-packages">Python Packages<a class="headerlink" href="#python-packages" title="Permanent link">¶</a></h3> <div class="highlight"><pre><code><a id="__codelineno-21-1" name="__codelineno-21-1" href="#__codelineno-21-1"></a># pyproject.toml additions <a id="__codelineno-21-2" name="__codelineno-21-2" href="#__codelineno-21-2"></a> <a id="__codelineno-21-3" name="__codelineno-21-3" href="#__codelineno-21-3"></a>[project.dependencies] <a id="__codelineno-21-4" name="__codelineno-21-4" href="#__codelineno-21-4"></a># Phase 1 <a id="__codelineno-21-5" name="__codelineno-21-5" href="#__codelineno-21-5"></a>spacy = "^3.7.0" <a id="__codelineno-21-6" name="__codelineno-21-6" href="#__codelineno-21-6"></a>networkx = "^3.2" <a id="__codelineno-21-7" name="__codelineno-21-7" href="#__codelineno-21-7"></a> <a id="__codelineno-21-8" name="__codelineno-21-8" href="#__codelineno-21-8"></a># Phase 3 <a id="__codelineno-21-9" name="__codelineno-21-9" href="#__codelineno-21-9"></a>transformers = "^4.35.0" # For zero-shot classification <a id="__codelineno-21-10" name="__codelineno-21-10" href="#__codelineno-21-10"></a>torch = "^2.1.0" # Or tensorflow <a id="__codelineno-21-11" name="__codelineno-21-11" href="#__codelineno-21-11"></a>scikit-learn = "^1.3.0" # For similarity calculations <a id="__codelineno-21-12" name="__codelineno-21-12" href="#__codelineno-21-12"></a> <a id="__codelineno-21-13" name="__codelineno-21-13" href="#__codelineno-21-13"></a>[project.optional-dependencies] <a id="__codelineno-21-14" name="__codelineno-21-14" href="#__codelineno-21-14"></a>activation = [ <a id="__codelineno-21-15" name="__codelineno-21-15" href="#__codelineno-21-15"></a> "spacy>=3.7.0", <a id="__codelineno-21-16" name="__codelineno-21-16" href="#__codelineno-21-16"></a> "networkx>=3.2", <a id="__codelineno-21-17" name="__codelineno-21-17" href="#__codelineno-21-17"></a> "transformers>=4.35.0", <a id="__codelineno-21-18" name="__codelineno-21-18" href="#__codelineno-21-18"></a> "torch>=2.1.0", <a id="__codelineno-21-19" name="__codelineno-21-19" href="#__codelineno-21-19"></a>] </code></pre></div> Model Downloads: <div class="highlight"><pre><code><a id="__codelineno-22-1" name="__codelineno-22-1" href="#__codelineno-22-1"></a># Phase 1 <a id="__codelineno-22-2" name="__codelineno-22-2" href="#__codelineno-22-2"></a>python -m spacy download en_core_web_sm # 17MB <a id="__codelineno-22-3" name="__codelineno-22-3" href="#__codelineno-22-3"></a> <a id="__codelineno-22-4" name="__codelineno-22-4" href="#__codelineno-22-4"></a># Phase 3 <a id="__codelineno-22-5" name="__codelineno-22-5" href="#__codelineno-22-5"></a># facebook/bart-large-mnli automatically downloaded by transformers (~1.6GB) </code></pre></div> <hr /> <h2 id="future-enhancements">Future Enhancements<a class="headerlink" href="#future-enhancements" title="Permanent link">¶</a></h2> <h3 id="short-term-next-6-months">Short-Term (Next 6 Months)<a class="headerlink" href="#short-term-next-6-months" title="Permanent link">¶</a></h3> 1. Activation Visualization - Export activation graph to Graphviz/D3.js - Show activation paths in UI - Debug activation patterns 2. Personalized Activation Parameters - Learn optimal decay rates per user - Adaptive hop count based on graph density - User-specific category taxonomies 3. Temporal Activation Patterns - Time-of-day aware activation - Seasonal/periodic memory patterns - Event-based activation triggers <h3 id="medium-term-6-12-months">Medium-Term (6-12 Months)<a class="headerlink" href="#medium-term-6-12-months" title="Permanent link">¶</a></h3> 4. Multi-Agent Spreading Activation - Shared memory graphs across agents - Collaborative activation (multiple agents activating same memory) - Agent-specific activation weights 5. Explanation Generation - Natural language explanations for why memories activated - "I remember this because you mentioned X and it relates to Y" - Transparency for user trust 6. Active Learning for Relations - User feedback on relation quality - Automatic relation type inference (beyond "related_to", "similar_to") - Reinforcement learning for optimal graph structure <h3 id="long-term-12-months">Long-Term (12+ Months)<a class="headerlink" href="#long-term-12-months" title="Permanent link">¶</a></h3> 7. Neuromorphic Activation - Spiking neural network-inspired activation - Continuous activation (not just on-demand) - Background memory consolidation 8. Cross-Modal Activation - Activate memories from images, audio, video - Multi-modal embeddings - Sensory-triggered recall 9. Metacognitive Activation - LLM self-reflection on activated memories - "Why did I remember this? Is it relevant?" - Confidence scores for activations <hr /> <h2 id="risks-mitigations">Risks & Mitigations<a class="headerlink" href="#risks-mitigations" title="Permanent link">¶</a></h2> <h3 id="risk-1-over-activation-too-many-memories">Risk 1: Over-Activation (Too Many Memories)<a class="headerlink" href="#risk-1-over-activation-too-many-memories" title="Permanent link">¶</a></h3> Impact: Medium - Context overload, slower LLM processing Mitigation: - Conservative threshold (0.15 default) - Limit max_results (5-10 default) - Decay activation strength with hops - User feedback: "Was this relevant?" <h3 id="risk-2-under-activation-missing-relevant-memories">Risk 2: Under-Activation (Missing Relevant Memories)<a class="headerlink" href="#risk-2-under-activation-missing-relevant-memories" title="Permanent link">¶</a></h3> Impact: High - Defeats purpose of natural activation Mitigation: - Lower threshold for testing (0.10) - Increase max hops (4-5) - Fallback to explicit search if activation returns <3 memories - Blend with review candidates (danger zone memories) <h3 id="risk-3-graph-sparsity-insufficient-relations">Risk 3: Graph Sparsity (Insufficient Relations)<a class="headerlink" href="#risk-3-graph-sparsity-insufficient-relations" title="Permanent link">¶</a></h3> Impact: Medium - Activation can't spread if no relations exist Mitigation: - Automatic relation creation (Phase 3) - Seed graph with common relations - Entity-based activation (doesn't require relations) - Encourage manual relation creation through UI <h3 id="risk-4-latency-from-graph-traversal">Risk 4: Latency from Graph Traversal<a class="headerlink" href="#risk-4-latency-from-graph-traversal" title="Permanent link">¶</a></h3> Impact: Low - Could slow conversation if >200ms Mitigation: - In-memory graph (NetworkX) for fast traversal - Limit max hops (3 default) - Async processing (don't block LLM response) - Cache activation results for similar queries <h3 id="risk-5-category-inference-accuracy">Risk 5: Category Inference Accuracy<a class="headerlink" href="#risk-5-category-inference-accuracy" title="Permanent link">¶</a></h3> Impact: Low - Wrong categories reduce multi-dimensional search quality Mitigation: - Zero-shot classification (no training required) - Conservative confidence threshold (0.5) - User feedback loop: Accept/reject category suggestions - Manual category override option <hr /> <h2 id="timeline-summary">Timeline Summary<a class="headerlink" href="#timeline-summary" title="Permanent link">¶</a></h2> <table> <thead> <tr> <th>Phase</th> <th>Duration</th> <th>Components</th> <th>Expected Impact</th> </tr> </thead> <tbody> <tr> <td>Phase 1</td> <td>3 weeks</td> <td>Spreading activation, trigger detection, hybrid scoring</td> <td>2-3x improvement in relevant memory retrieval</td> </tr> <tr> <td>Phase 2</td> <td>2 weeks</td> <td>MCP tool integration, API exposure</td> <td>Usable natural activation in conversations</td> </tr> <tr> <td>Phase 3</td> <td>3 weeks</td> <td>Category inference, auto-relations, multi-dimensional search</td> <td>3-4x improvement, graph density increase</td> </tr> <tr> <td>Phase 4</td> <td>2 weeks</td> <td>Performance tuning, user testing, documentation</td> <td>Production-ready system</td> </tr> <tr> <td>Total</td> <td>10 weeks</td> <td>Complete natural language activation system</td> <td>3-4x overall improvement</td> </tr> </tbody> </table> <hr /> <h2 id="conclusion">Conclusion<a class="headerlink" href="#conclusion" title="Permanent link">¶</a></h2> This implementation plan transforms cortexgraph from explicit memory retrieval to natural conversational activation, leveraging cutting-edge research from 2024-2025 while building on cortexgraph's unique temporal memory foundations. Key Innovations: 1. Hybrid spreading activation + temporal decay - Unique combination not seen in existing systems 2. Multi-dimensional search - Inspired by IMDMR, adapted for temporal memory 3. Automatic relation creation - Graph density improvement without manual effort 4. Integration with natural spaced repetition - Activated memories automatically reinforce Expected Outcome: Conversational AI that remembers context naturally, achieving 3-4x improvement in relevant memory retrieval compared to explicit search baseline. <hr /> <h2 id="references">References<a class="headerlink" href="#references" title="Permanent link">¶</a></h2> <h3 id="academic-papers-2024-2025">Academic Papers (2024-2025)<a class="headerlink" href="#academic-papers-2024-2025" title="Permanent link">¶</a></h3> <ol> <li>IMDMR (arxiv:2511.05495v1) - Multi-dimensional memory retrieval</li> <li>SpreadPy (arxiv:2507.09628) - Spreading activation library</li> <li>Mem0 (arxiv:2504.19413) - Production-ready scalable memory</li> <li>A-MEM (arxiv:2502.12110) - Agentic memory with Zettelkasten</li> <li>MIRIX (arxiv:2507.07957v1) - Multi-agent memory system</li> <li>SynapticRAG (arxiv:2410.13553) - Temporal memory retrieval</li> <li>Semantic Network Model (arxiv:2301.11709v1) - Spreading activation for comprehension</li> </ol> <h3 id="industry-systems">Industry Systems<a class="headerlink" href="#industry-systems" title="Permanent link">¶</a></h3> <ul> <li>Mem0: github.com/mem0ai/mem0</li> <li>Memori: github.com/GibsonAI/Memori</li> <li>spaCy: spacy.io</li> <li>NetworkX: networkx.org</li> <li>Transformers (Hugging Face): huggingface.co/transformers</li> </ul> <h3 id="cortexgraph-documentation">CortexGraph Documentation<a class="headerlink" href="#cortexgraph-documentation" title="Permanent link">¶</a></h3> <ul> <li>conversational-activation-plan.md - Preprocessing for save detection</li> <li>architecture.md - Temporal decay and natural spaced repetition</li> <li>graph_features.md - Knowledge graph structure</li> <li>api.md - Existing MCP tools</li> </ul> <hr /> Document Version: 1.0 Last Updated: 2025-11-14 Author: Claude (Sonnet 4.5) Branch: <code>feat/natural-language-activation</code> Next Review: After Phase 1 completion </article> </div> <script>var target=document.getElementById(location.hash.slice(1));target&&target.name&&(target.checked=target.name.startsWith("__tabbed_"))</script> </div> <button type="button" class="md-top md-icon" data-md-component="top" hidden> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M13 20h-2V8l-5.5 5.5-1.42-1.42L12 4.16l7.92 7.92-1.42 1.42L13 8z"/></svg> Back to top </button> </main> <footer class="md-footer"> <div class="md-footer-meta md-typeset"> <div class="md-footer-meta__inner md-grid"> <div class="md-copyright"> Made with <a href="https://squidfunk.github.io/mkdocs-material/" target="_blank" rel="noopener"> Material for MkDocs </a> </div> <div class="md-social"> <a href="https://github.com/prefrontal-systems/cortexgraph" target="_blank" rel="noopener" title="github.com" class="md-social__link"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg> </a> <a href="https://github.com/prefrontal-systems/cortexgraph/discussions" target="_blank" rel="noopener" title="github.com" class="md-social__link"> <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 576 512"><path d="M384 144c0 97.2-86 176-192 176-26.7 0-52.1-5-75.2-14l-81.6 43.2c-9.3 4.9-20.7 3.2-28.2-4.2s-9.2-18.9-4.2-28.2l35.6-67.2C14.3 220.2 0 183.6 0 144 0 46.8 86-32 192-32s192 78.8 192 176m0 368c-94.1 0-172.4-62.1-188.8-144 120-1.5 224.3-86.9 235.8-202.7 83.3 19.2 145 88.3 145 170.7 0 39.6-14.3 76.2-38.4 105.6l35.6 67.2c4.9 9.3 3.2 20.7-4.2 28.2s-18.9 9.2-28.2 4.2L459.2 498c-23.1 9-48.5 14-75.2 14"/></svg> </a> </div> </div> </div> </footer> </div> <div class="md-dialog" data-md-component="dialog"> <div class="md-dialog__inner md-typeset"></div> </div> <script id="__config" type="application/json">{"annotate": null, "base": "..", "features": ["navigation.instant", "navigation.tracking", "navigation.tabs", "navigation.sections", "navigation.expand", "navigation.top", "search.suggest", "search.highlight", "content.code.copy", "content.code.annotate", "content.action.edit"], "search": "../assets/javascripts/workers/search.7a47a382.min.js", "tags": null, "translations": {"clipboard.copied": "Copied to clipboard", "clipboard.copy": "Copy to clipboard", "search.result.more.one": "1 more on this page", "search.result.more.other": "# more on this page", "search.result.none": "No matching documents", "search.result.one": "1 matching document", "search.result.other": "# matching documents", "search.result.placeholder": "Type to start searching", "search.result.term.missing": "Missing", "select.version": "Select version"}, "version": null}</script> <script src="../assets/javascripts/bundle.e71a0d61.min.js"></script> </body> </html>

Latest Blog Posts

What Is Context Bloat in MCP?
By Om-Shree-0709 on December 16, 2025.
mcp
Context Bloat
MCP Moves to the Linux Foundation: Neutral Stewardship for Agentic Infrastructure
By Om-Shree-0709 on December 15, 2025.
mcp
anthropic
Linux Foundation
Code Execution with MCP: Architecting Agentic Efficiency
By Om-Shree-0709 on December 14, 2025.
mcp
Token bloat

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/prefrontalsys/mnemex'

If you have feedback or need assistance with the MCP directory API, please join our Discord server