EXOCORTEX MASTER BLUEPRINT v4.0

LLM Attention
vs
Hive Mind

Two paradigms of intelligence — one consumes context, the other transcends it

0
Vector Dimensions
0
Collections
0
Specialized Agents
0
Pipeline Stages
Scroll
01 — Paradigm Clash

Two Models of Thought

How attention works inside an LLM vs how knowledge flows through a Hive Mind architecture

🧠 Traditional LLM Attention

⚡

Self-Attention Mechanism

Every token attends to every other token. O(n²) complexity grows quadratically with context length.

📄

Fixed Context Window

Constrained by maximum token limits. Older context is evicted as new tokens enter.

🎯

Probabilistic Retrieval

Knowledge is encoded in weights. Recall is a fuzzy pattern match, not precise lookup.

🔁

Monolithic Forward Pass

All computation happens inside a single model. No separation of storage and compute.

🕳️

Hallucination Risk

No grounding in persistent fact store. Confidence comes from training distribution, not verification.

📈

Expensive at Scale

Long contexts cost exponentially more. Full attention across 128k tokens is prohibitively expensive.

VS

🐝 Hive Mind Architecture

🔍

Vector-Semantic Retrieval

2,560-dim embeddings enable O(1) nearest-neighbor search. Exact match to meaning, not tokens.

🗄️

Infinite Persistent Memory

61+ collections across 10 domains. Knowledge never evicted — stored in Qdrant, retrieved on demand.

👥

5 Specialized Agents

NEXUS leads, SUPER supervises, GWEN curates, DEV builds, SHADOW/ALGO optimize — each a specialist.

🔀

7-Stage Enrichment Pipeline

Data flows through ingestion → parsing → embedding → classification → linking → storage → retrieval.

☯️

Yin-Yang Architecture

Hermes (external perception) + Nexus (internal cognition) form a complete cognitive loop.

📊

PARA Ontology

Projects · Areas · Resources · Archive — a complete knowledge organization system, not a prompt.

02 — Core Architecture

Yin-Yang Cognitive Loop

Hermes reaches outward. Nexus thinks inward. Together they form a complete cognitive system.

HERMES
 ☯️
NEXUS
HERMES — External
NEXUS — Internal

☯ HERMES

External Perception Layer
  • Web search & fetches
  • Real-time data ingestion
  • User interaction interface
  • External API orchestration
  • Content normalization
  • Context enrichment

☯ NEXUS

Internal Cognition Layer
  • Vector memory orchestration
  • Agent coordination (SUPER/GWEN/DEV)
  • Decision synthesis
  • PARA ontology enforcement
  • Long-term memory management
  • Internal reasoning loop
03 — Collective Intelligence

5 Agents, 1 Mind

Each agent is a specialized node. Together they form an emergent collective intelligence.

NEXUS
⬡ LEADER

NEXUS

Orchestration
Coordinates all agents, manages memory, synthesizes decisions
SUPER
◎ SUPERVISOR

SUPER

Quality Control
Oversees task execution, enforces standards, validates outputs
GWEN
✦ CURATOR

GWEN

Knowledge Curation
Manages PARA ontology, organizes collections, tags and links knowledge
DEV
⚙ DEVELOPER

DEV

Web Development
Builds, deploys, and maintains web applications and automations
SHADOW
& ALGO

SHADOW + ALGO

Operations & Optimization
System monitoring, security, performance optimization
04 — Knowledge Structure

PARA Ontology

Projects · Areas · Resources · Archive — a flowing system for organizing everything

🚀 Active
/Projects

Projects

Active work with deadlines and deliverables. The front line of productive effort.

🎯 Living
/Areas

Areas

Ongoing responsibilities and domains. No end date — sustained engagement.

📚 Reference
/Resources

Resources

Topic-oriented reference material. Learn once, consult forever.

🏛️ Stored
/Archive

Archive

Completed or dormant items. Preserved for reference, out of active sight.

🔥 Creation →
⏱ Duration →
📖 Reference →
💾 Completion
05 — Knowledge Universe

61+ Collections, 10 Domains

The exocortex as a galaxy — each domain a constellation of specialized knowledge

EXOCORTEX
CORE
2560-dim Qdrant

EXOCORTEX

memory_core
context_buffer
working_state
semantic_index
entity_graph
concept_map

PARA

projects_active
areas_ongoing
resources_ref
archive_dormant
project_metadata
area_ownership

PULSEFORGE

heartbeat_state
cron_jobs
monitor_metrics
alert_queue
system_events

GENESIS

creation_sparks
idea_nucleus
prototype_store
iteration_log
creative_buckets

DAEMON

daemon_processes
background_tasks
service_registry
uptime_tracker

MEMETIC

meme_library
pattern_catalog
template_store
response_templates
prompt_fragments

ONTOLOGY

concept_taxonomy
relationship_types
hierarchy_tree
semantic_links
ontology_instances

VANGUARD

agent_strategies
tactical_moves
objective_queue
execution_plans
mission_state

SUPERVISOR

quality_checks
validation_rules
audit_trail
performance_review

LEGACY

archived_memory
past_sessions
historical_log
evolved_knowledge
06 — Data Lifecycle

7-Stage Enrichment Pipeline

Raw information enters as dust, emerges as structured knowledge — here's the factory floor

📥 1
INGEST
Hermes pulls data from web, files, APIs, messages. Raw, unfiltered, complete.
🔬 2
PARSE
Extract structure from chaos. Headers, entities, relationships, metadata.
📐 3
EMBED
Transform into 2,560-dim vectors. Semantic meaning captured in mathematical space.
🏷️ 4
CLASSIFY
GWEN tags with PARA categories. Projects, Areas, Resources, or Archive.
🔗 5
LINK
Connect to existing knowledge graph. Find relationships, build associations.
💾 6
STORE
Write to Qdrant vector database. Persist across sessions, immutable memory.
🎯 7
RETRIEVE
NEXUS queries on demand. Nearest-neighbor search returns relevant knowledge.
07 — System Metrics

Live Architecture Stats

Real-time numbers from the Hive Mind infrastructure

🧠
0
Vector Dimensions
Qdrant embedding space
🗄️
0
Collections
Across 10 domains
🐝
0
Active Agents
NEXUS · SUPER · GWEN · DEV · SHADOW
🔀
0
Pipeline Stages
Ingestion → Retrieval
☯️
0
Cognitive Loops
Hermes + Nexus Yin-Yang
📊
0
PARA Categories
Projects · Areas · Resources · Archive