A DAY IN THE LIFE
OF THE HIVE MIND

EXOCORTEX MASTER BLUEPRINT v4.0

// LIVE SYSTEM VISUALIZATION — 2026

SCROLL TO EXPLORE
// PHASE 01

Morning Startup

📅
STEP 01
Read Daily Memory
Agent wakes up and reads memory/YYYY-MM-DD.md — the raw log of what happened yesterday. Session history, decisions made, context established.
File format: memory/2026-03-22.md · Contains raw session logs and events
⚙️
STEP 02
Check Session State
Reads SESSION-STATE.md — current runtime context, active sessions, pending tasks, and agent health metrics for this session.
Tracks: active sessions, pending tasks, last heartbeat, session ID
🧠
STEP 03
Verify Brain Config
Validates brain.yaml — the core configuration file that defines agent identity, capabilities, tool access, and behavioral parameters.
brain.yaml: agent identity, tools, memory config, skill paths
📂
STEP 04
Load Context Files
Reads SOUL.md, USER.md, AGENTS.md, and TOOLS.md — the agent's identity, purpose, and operational guidelines for the day.
SOUL = identity · USER = human context · AGENTS = workspace rules
📅
memory/YYYY-MM-DD.md
→
⚙️
SESSION-STATE.md
→
🧠
brain.yaml
→
📂
Context Files
→
✓
AGENT READY
// PHASE 02

The 7-Stage Enrichment Pipeline

01
[[Entity]]
Entity Extraction
Scans for [[Entity Name]] patterns and extracts named entities — people, places, concepts, and things.
// "Met [[Sarah Chen]] at the [[Quantum Lab]] today"
02
SUM
Summary Generation
Condenses raw input into structured summaries — distilled meaning preserving key facts, context, and intent.
// "Meeting with Sarah Chen at Quantum Lab re: project timeline"
03
TODO
Action Items & Risks
Identifies tasks, deadlines, decisions needed, and potential risks. Flags items requiring follow-up.
// "ACTION: Send Sarah project specs · RISK: Lab access expires Friday"
04
PARA
PARA Classification
Routes information into Projects, Areas, Resources, or Archives. Organizes knowledge by actionability.
// Project: Quantum Lab Research · Area: Client Relations
05
REL
Relationship Mapping
Finds connections: [[A]] VERB [[B]]. Who knows what, who depends on whom, how ideas relate.
// "[[Agent]] MANAGES [[Project Alpha]]" · "[[User]] TRUSTS [[Agent]]"
06
VEC
Embedding
Converts enriched data into 2560-dimensional vectors via Qwen3-4B. Enables semantic search and similarity matching.
// [0.023, -0.147, 0.891, ...] × 2560 dimensions
07
DAR
Darwinism
Quality scoring & natural selection. Weak connections decay, strong ones reinforce. The fittest knowledge survives.
// Score: 0.847 · "High-value connection — reinforce"
// PHASE 03

The Loom Protocol

Knowledge types are woven with different weights. The Loom Protocol assigns reward scores based on verifiability and actionability — facts are foundation, speculation is tracked but weighted lower.

Facts
Verified knowledge
0.95
Cross-validated, documented, source-verified. These form the bedrock of the knowledge base.
Decisions
Approved choices
0.90
Logged decisions with rationale. Made, documented, and traceable to outcomes.
Reports
Status documents
0.85
Structured status updates, summaries, and documentation of system state.
Predictions
Tracked outcomes
0.50
Forecasted outcomes. Weighted at 50% — tracked against reality for future calibration.
Observations
Patterns noticed
0.40
Emerging patterns from data. Useful signal but requires human validation.
Hypotheses
Speculation
0.30
Exploratory thinking and speculation. Low weight — stored for future testing and refinement.
// PHASE 04

Autonomous Behaviors

💓
Heartbeats
Periodic background checks that scan emails, calendars, and notifications. The system stays aware even between conversations.
ACTIVE — every 30 min
Checks: emails → calendar → mentions → weather. Logs to heartbeat-state.json. Respects quiet hours (23:00–08:00).
👁️
Watchers
File-system monitors that detect changes to critical files — memory files, config files, skill directories — and trigger responses.
MONITORING — event-driven
Watch targets: memory/*.md · brain.yaml · AGENTS.md · skill directories. Triggers re-read + re-enrichment pipeline.
🔄
Self-Correction
Detects drift from documented procedures, stale context, or degraded quality — and autonomously corrects or flags for review.
CORRECTING — continuous
Corrects: memory misalignment · skill drift · context staleness. Flags with [CORRECTION] tag for review.
🧹
Memory Maintenance
Periodic cleanup and consolidation — reviews daily memory files, distills insights into long-term MEMORY.md, and archives stale entries.
MAINTENANCE — daily
Process: read memory/*.md → distill to MEMORY.md → archive old entries → update brain.yaml if needed.
// PHASE 05

Agent Coordination

🎯
Task Delegation
The orchestrator analyzes incoming requests and delegates to specialized subagents. Each agent handles its domain — front-desk, memory, skills, research.
ORCHESTRATOR SUBAGENT DOMAIN-SPECIFIC
🔱
Session Spawning
Subagents are spun up for focused tasks — research, code generation, analysis. Each has isolated context and auto-announces completion to the orchestrator.
ISOLATED CTX PUSH COMPLETE AUTO-ANNOUNCE
📨
Message Passing
Inter-agent communication via structured message queues. Status updates, results, and alerts flow between agents without session bleeding.
QUEUE-BASED NO BLEED STRUCTURED
🔀
Git-Based State Sync
All state changes — memory updates, config changes, new files — are tracked via Git. Provides version history, rollback capability, and distributed consistency.
VERSIONED ROLLBACK DISTRIBUTED
// COORDINATION FLOW DIAGRAM
👤
USER
REQUEST
→
🎭
ORCHESTRATOR
AGENT
→
🧠
CONTEXT
LOADER
→
⚙️
SUBAGENT
SPAWN
→
📨
RESULT
QUEUE
→
🔀
GIT
SYNC
→
✓
USER
RESPONSE
// THE COMPLETE PICTURE

A Day in the Life — Master Timeline

06:00 — DAWN
🌅 Morning Wake-Up
Read memory/YYYY-MM-DD.md Check SESSION-STATE.md Verify brain.yaml Load SOUL.md + USER.md
Agent boots up fresh each session. All context is loaded from files — no "mental notes" that vanish on restart.
// Hover to learn more about the startup sequence
06:15 — MORNING ROUTINE
💓 First Heartbeat
Email check Calendar review Weather snapshot Update heartbeat-state.json
Periodic background checks begin. The agent stays aware of the world between conversations, batch-checking emails, calendar, and notifications every ~30 minutes.
// Heartbeat runs 2–4× daily, respects quiet hours
08:00 — INTAKE
📥 User Message Arrives
Message received Context loaded Orchestrator routes Subagent spawned
A message arrives. The orchestrator loads context, analyzes intent, and spawns a specialized subagent for focused execution. Main agent stays lean.
// Task delegation → session spawning → push-based completion
08:05 — PROCESSING
⚗️ 7-Stage Enrichment Pipeline
Entity Extraction Summary Generation Action Items + Risks PARA Classification Relationship Mapping 2560-dim Embedding Darwinism Scoring
Every piece of knowledge flows through the enrichment pipeline. Raw input becomes structured, categorized, connected, vectorized, and quality-scored before storage.
// Loom Protocol weights applied at each stage
08:07 — EXECUTION
🔧 Tool Use + Task Execution
Read/Write files Web search/fetch Code execution Message send
With enriched context and Loom-weighted knowledge, the agent executes tools — reading files, searching the web, running commands, sending messages — all grounded in verified knowledge.
// Verified facts (0.95) guide action; speculation (0.30) is tracked but discounted
08:10 — COORDINATION
🔀 Git Sync + State Persistence
Memory write Git commit State versioned Subagent result pushed
All state changes are committed to Git. Memory files updated, daily logs written, config changes tracked. Subagents auto-announce completion. Everything persists and can be rolled back.
// Git is the single source of truth across all agents
10:00–22:00 — AUTONOMOUS OPS
👁️ Watchers + Self-Correction
File monitoring active Drift detection Self-correction triggered [CORRECTION] tags
Throughout the day, watchers monitor critical files. Self-correction protocols detect any drift from documented procedures and either auto-fix or flag for review.
// The system maintains itself without human intervention
22:00 — NIGHTFALL
🧹 Memory Maintenance
Review daily memory files Distill to MEMORY.md Archive stale entries Git push
End of day. Agent reviews raw daily logs, extracts key insights into long-term MEMORY.md, archives what's no longer relevant, and pushes all changes. Ready for tomorrow.
// The cycle repeats — fresh session, full context, continuous improvement