Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<p align="center"><img src="./icons/knowledge-system-architecture.png" alt="Madeinoz Knowledge System Architecture"></p>
Knowledge
Persistent personal knowledge management system powered by Graphiti knowledge graph - automatically extracts entities, relationships, and temporal context from conversations and documents.
Changelog
See CHANGELOG.md for full version history.
Documentation
[View Full Documentation](https://madeinoz67.github.io/madeinoz-knowledge-system/) - Complete guides, architecture, and reference.
| Topic | Description |
|---|---|
| Getting Started | Installation and quick start guide |
| Configuration | Environment variables and settings |
| Architecture | System design and components |
| Troubleshooting | Common issues and solutions |
| Developer Notes | Contributing and development |
Installation
See INSTALL.md for complete installation instructions, performance benchmarks, and VERIFY.md for verification checklist.
Features
- Automatic Entity Extraction - LLM-powered extraction of people, organizations, concepts, and more
- Relationship Mapping - Automatically discovers connections between entities
- Semantic Search - Find knowledge using natural language, not just keywords
- Investigative Search - Find entity with all connected relationships in a single query (configurable depth 1-3 hops)
- Memory Decay Scoring - Automatic memory prioritization with importance/stability classification
- Weighted Search - Results ranked by semantic relevance, recency, and importance
- Lifecycle Management - Automated memory transitions (ACTIVE → DORMANT → ARCHIVED → EXPIRED)
- Prometheus Metrics - Token usage, API costs, cache statistics, and memory health metrics
- Automated Maintenance - Scheduled cleanup of expired memories
- Grafana Dashboards - Visualize knowledge, token usage, graph stats and memory health
- Temporal Tracking - Know when knowledge was captured and how it evolves
- Memory Sync - Auto-syncs learnings from PAI Memory System
- OSINT/CTI Ontology - Custom entity types for threat intelligence (ThreatActor, Malware, Vulnerability, Indicator, etc.) with STIX 2.1 import support
Usage
The skill triggers automatically based on natural language:
| Say This | Action |
|---|---|
| "remember that X" | Capture knowledge with entity extraction |
| "what do I know about X" | Semantic search for related entities |
| "how are X and Y related" | Find relationships between concepts |
| "what did I learn today" | Temporal search - filter by date |
| "recent learnings" | Retrieve recent knowledge additions |
| "knowledge status" | Check system health |
Temporal Search
Filter search results by date with --since and --until:
# Today's knowledge
bun run tools/knowledge-cli.ts search_nodes "topic" --since today
# Last 7 days
bun run tools/knowledge-cli.ts search_facts "decisions" --since 7d
# Date range
bun run tools/knowledge-cli.ts search_nodes "project" --since 2026-01-01 --until 2026-01-15Date formats: today, yesterday, 7d, 1w, 1m, or ISO dates (2026-01-26)
Weighted Search (Low-Cost)
Rank results by semantic relevance (60%) + recency (25%) + importance (15%) using the --weighted flag:
# Weighted search - prioritizes important, recent, relevant knowledge
bun run tools/knowledge-cli.ts search_nodes "topic" --weightedCost benefit: Weighted scoring uses already-computed embeddings and metadata — no additional LLM calls. Works with any embedding model including free/local options like Ollama, Trinity, or gpt-4o-mini.
Output includes:
- 📊 Overall score (0-1)
- S: Semantic similarity
- R: Recency score
- I: Importance score
- Lifecycle state (ACTIVE/DORMANT/ARCHIVED)
- Importance/Stability ratings (1-5)
What's Included
| Component | Purpose |
|---|---|
SKILL.md | PAI skill with intent-based routing |
src/skills/workflows/ | 8 workflows (Capture, Search, SearchByDate, Facts, Recent, Status, Clear, BulkImport) |
src/skills/tools/ | Server management scripts (start, stop, status, logs) |
src/hooks/ | Memory sync hook for automatic knowledge capture |
docker/ | Docker/Podman compose files for Neo4j and FalkorDB |
Database Backends
| Backend | Web UI | Best For |
|---|---|---|
| Neo4j (default and recommended) | <http://localhost:7474> | Rich queries, special character handling |
| FalkorDB (experimental) | <http://localhost:3000> | Simple setup, lower resources |
For AI Agents
This is a PAI Pack - a complete, self-contained module for Personal AI Infrastructure:
- Read the entire README to understand what you're installing
- Follow INSTALL.md step-by-step
- Complete ALL verification checks in VERIFY.md
- If any step fails, STOP and troubleshoot before continuing
Credits
- Knowledge graph engine: Graphiti by Zep AI
- Graph databases: Neo4j, FalkorDB
- Built for: Personal AI Infrastructure (PAI)
See full Acknowledgments for credits to the community and research that inspired this system.
Related
- PAI Memory System - Auto-syncs learnings to knowledge graph
- PAI Research Skill - Capture research findings
For detailed documentation, visit the [full docs](https://madeinoz67.github.io/madeinoz-knowledge-system/).