Solo architect of self-modifying AI systems.
Building agents that read their own code, fix their own bugs, and evolve their own architecture.
I design and build Genesis Agent — a self-aware, self-modifying cognitive AI agent with a 12-phase boot system, hexagonal architecture, and organism substrate. It doesn't just use LLMs — it wraps them in 247 modules of self-verification, self-repair, causal reasoning, and autonomous planning.
247 source files · ~73k LOC · 4354 tests · 152 services
12 boot phases · 130/130 architectural fitness · coverage 82/76/80
16 hash-locked safety files · 242 late-bindings · 385 event types
Crash-safe sessions · Frontier-based memory · Heuristic self-scoring
Runs on Claude, GPT-4, DeepSeek, Kimi, or local models via Ollama
┌─────────────────────────────────────────────────────────────┐
│ GENESIS AGENT v7.2.1 — Cognitive Architecture │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │Foundation │→ │Intellig. │→ │Capabilit.│→ │Planning │ │
│ │ Phase 1 │ │ Phase 2 │ │ Phase 3 │ │ Phase 4 │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │Hexagonal │→ │Autonomy │→ │Organism │→ │Revolution│ │
│ │ Phase 5 │ │ Phase 6 │ │ Phase 7 │ │ Phase 8 │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │Cognitive │→ │ Agency │→ │Extended │→ │ Hybrid │ │
│ │ Phase 9 │ │Phase 10 │ │Phase 11 │ │Phase 12 │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
└─────────────────────────────────────────────────────────────┘
| System | Purpose |
|---|---|
| CausalAnnotation | Deterministic cause-effect tracking without LLM |
| InferenceEngine | Rule-based reasoning — answers without calling the model |
| GoalSynthesizer | Autonomous goal generation from empirical weaknesses |
| StructuralAbstraction | Cross-context pattern learning from concrete experiences |
| CognitiveWorkspace | 9-slot working memory (Global Workspace Theory) |
| DreamCycle | Offline memory consolidation + insight generation |
| FrontierNode | Graph-based focus anchor — "what's important now" |
| SessionPersistence | Crash-safe checkpoints + heuristic session scoring |
| UnifiedMemory | Cross-store recall with Jaccard similarity boosting |
| ColonyOrchestrator | Multi-agent goal decomposition + merge |
| SelfModificationPipeline | Sandboxed code changes with rollback |
| PreservationInvariants | Hash-locked semantic safety rules |
| Version | Highlight |
|---|---|
| v7.2.1 | Solid Ground II — deep audit, 8 silent bugs fixed, 16 hash-locked files |
| v7.2.0 | Self-Define — Genesis writes its own identity from deterministic data |
| v7.1.9 | Binding Visibility — expectedActive flag makes silent failures visible at boot |
| v7.1.5 | Emotional Continuity — cross-layer EmotionalFrontier (Phase 8 bridge) |
| v7.1.4 | Session-Aware Memory — crash-safe checkpoints, frontier node, cross-referencing. Inspired by neo.mjs Memory Core. |
| v7.1.3 | Real rollback via SnapshotManager, fitness 130/130, coverage push |
Solo developer. I ship iteratively — build, verify, release. Every module has tests. Every event has a payload schema. Every release runs through architectural fitness scoring. The codebase maintains itself.
Stack: JavaScript · Node.js · Electron · MCP Protocol · Ollama · Anthropic · OpenAI · DeepSeek · Kimi
Germany · Building autonomous systems that understand their own architecture.
