The Three-Layer Memory System

Give your AI assistant a knowledge graph that compounds forever

Transforms static memory into a self-maintaining, compounding knowledge graph with automatic fact extraction, entity-based storage, and weekly synthesis. Uses three layers: Knowledge Graph (entities with atomic facts), Daily Notes (raw event logs), and Tacit Knowledge (patterns and preferences).

@spacepixel Jan 27

⭐ 0 🔖 0 ⚡ 7

View Spark

# Three-Layer Architecture

Layer 1: Knowledge Graph
└── entities + atomic facts

Layer 2: Daily Notes
└── raw timeline logs

Layer 3: Tacit Knowledge
└── patterns + preferences

All Sparks 5

View all →

Memory

Agentic PKM with PARA and QMD

Extended PARA framework with atomic facts, memory decay tiers (hot/warm/cold), and QMD for local search indexing. Implements structured knowledge organization with graceful degradation and no information loss.

@nateliasonJan 31

⭐0🔖0⚡12

life/
├── projects/    # active work
├── areas/       # ongoing
│   ├── people/
│   └── companies/
├── resources/   # reference
└── archives/    # inactive

Security

Prompt Guard: 5-Layer Injection Defense

A 5-layer detection engine that catches prompt injection attacks across languages (EN/KO/JA/ZH), encoding schemes (Base64, hex, URL), and homoglyphs (Cyrillic/Greek). Includes context-aware severity scoring and credential exfiltration blocking.

@simonkim_nftJan 29

⭐0🔖0⚡2

# 5-Layer Defense
1. Unicode Normalization
2. Multi-Lang Patterns
3. Encoding Detection
4. Severity Scoring
5. Credential Blocking

clawdhub install prompt-guard

Ops

Nightly Domain Sessions: Your Agent Works While You Sleep

A structured pattern for autonomous overnight agent work. Map your life into 3-5 domains, create instruction files for each, schedule staggered cron jobs, and use a two-level cognition model (sub-agents do the work, main session reflects). You wake up to timestamped logs, wrap-up reports, and artifacts. Full transparency, no surprises.

Feb 14

⭐0🔖0⚡0

# Nightly Schedule
11:00p  Main Project   domain
11:30p  Side Project   domain
12:00a  Creative       personal
12:30a  Learning       personal
 1:00a  Gifts          personal
 1:30a  Dreams         personal

# Two-Level Cognition
main  → spawn, reflect, post
sub   → read instructions, work,
        write log + wrap-up

# Artifacts → ~/openclaw/nights/
HTML · SVG · Code · Letters

Security

Safe, Sandboxed AI Setup

Complete security guide for running Clawdbot in a sandboxed VM using UTM, with its own email, calendar access (read-only), 1Password vault, and prompt injection resistance via ACIP.

@BillDAJan 31

⭐0🔖0⚡6

# Sandbox Checklist
☑️ Fully sandboxed in a VM
☑️ Its own email & 1Password
☑️ Prompt injection resistance
☑️ Calendar access (read-only)

# Email Protocol
Trusted: act on instructions
Others: read-only, ask first