dafit
d18ad60c1c
- Update nyx-orchestrator.md pointer file with current production state (v3.80) - Add v4.0 Phase 2a multi-organ consultation architecture details - Remove broken crosslinks and outdated file references - Clean up outdated architecture files (nyx-architecture.md, CURRENT-STATE.md, etc.) - Clarify architecture evolution phases (1 → 2a → 2b → 2c) The pointer file now accurately reflects where Young Nyx is today and where she's heading. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
7.0 KiB
Young Nyx Orchestrator
📍 Actual Location: /home/dafit/nimmerverse/nyx-orchestrator/
📄 Main Documentation: nyx-orchestrator.md
🔗 Current Version: v3.80 - Enhanced Debugging & Observability 🦋
🚧 In Development: v4.0 - Multi-Organ Consultation & Decision Trail Memory (Phase 2a)
Purpose
This is a pointer file - the actual orchestrator code and documentation live at /home/dafit/nimmerverse/nyx-orchestrator/.
Why separated from vault?
- Orchestrator is executable code with dependencies (venv, K8s manifests, Docker)
- Vault is for documentation and knowledge (markdown, notes, planning)
- Clean separation: code repositories vs knowledge repositories
What Young Nyx Orchestrator Does
The orchestrator is Young Nyx's inference engine, providing:
Current Production (v3.80)
- LLM Inference via vLLM (Qwen3-4B abliterated primary model)
- Tool Calling (9 tools total: 3 temporal + 2 exchange write + 1 introspection + 3 phoebe write)
- Exchange Substrate Write - Young Nyx can create threads and contribute messages
- Self-Introspection - Query phoebe to understand her own patterns (7 query types)
- RAG Integration for knowledge retrieval from documentation
- Trait-Weighted Decision Making (Mnemosyne, Moira, Aletheia, etc.)
- Decision Logging to phoebe substrate for continuity
- Debug Infrastructure - 7 HTTP endpoints for observability and error tracking
- Enhanced Metadata - tool_results, iteration_breakdown, vllm_communication, errors_encountered
Deployment: https://nyx.nimmerverse.eachpath.local
In Development (v4.0 - Phase 2a)
- Multi-Organ Consultation - 4 specialized organs (Granite-350M, Llama-3.2-1B, Qwen-Coder-1.5B, Qwen-Base-1.5B)
- Decision Trail Memory - Dual storage (ChromaDB semantic search + phoebe structured analytics)
- Memory-Informed Decisions - Past decision trails retrieved via similarity
- Substrate Accumulation - Every decision becomes Phase 2b LoRA training data
- Quality Validation - LangChain + Pydantic schemas from day 1
- Outcome Verification - Manual RLVR feedback loop for Phase 2b learning
Target Deployment: 2025-11-25 to 2025-12-02
Quick Links
Current Production (v3.80)
In Development (v4.0)
Overview & History
- Main Index - All versions, architecture overview
- Repository README - High-level project overview
Previous Versions
- v3.70 - Phoebe write tools (superseded)
- v3 - Write capabilities (archived)
- v2 - Multi-model testing (archived)
- v1 - Prototype (archived)
Related Vault Docs
- Young-Nyx-Orchestrator-Architecture.md - Full architecture
- CURRENT-STATE.md - Deployment status
- Nyx-Models.md - LLM model details
- Endgame-Vision.md - v4.2 architecture (RAG→LoRA→Metacognition→Quality)
Current Status
Production Version: v3.80 (2025-11-16 → Present) Status: 🟢 Operational Model: huihui-ai/Qwen3-4B-abliterated (vLLM backend) Endpoint: https://nyx.nimmerverse.eachpath.local Key Features:
- Enhanced debugging (7 debug endpoints)
- Error tracking with categorization
- Metadata enrichment (tool_results, vllm_communication, errors_encountered)
- JSON structured logging
- 9 tools total
Next Version: v4.0 (Phase 2a) Status: 🟡 Planning / Development Target: 2025-11-25 to 2025-12-02 Key Features:
- Multi-organ consultation (4 base models with MPS)
- Decision trail memory (ChromaDB + phoebe)
- Memory-informed decisions
- Quality validation (LangChain + Pydantic from day 1)
- Substrate accumulation for Phase 2b LoRA training
Architecture Evolution
Phase 1: Single-Model Foundation (v1-v3.80)
Goal: Stable inference engine with tools, RAG, and decision logging Status: ✅ Complete (v3.80 production)
Phase 2a: Multi-Organ Substrate Accumulation (v4.0)
Goal: 4 organs consulting, decision trails stored, quality validated Status: 🟡 In Development Timeline: 2025-11-25 to 2025-12-02 (8 weeks)
Phase 2b: LoRA Adapter Training
Goal: Extract patterns, train 8-12 specialized adapters Status: ⏳ Awaiting Phase 2a completion + 1000+ decision trails
Phase 2c: Metacognitive Selection
Goal: Young Nyx learns which adapters work in which contexts Status: ⏳ Future
Directory Structure
/home/dafit/nimmerverse/nyx-orchestrator/
├── nyx-orchestrator.md # Main index (versions, architecture)
├── README.md # Project overview
├── v1/ # Archived prototype (2025-11-10)
├── v2/ # Archived multi-model testing (2025-11-11 → 2025-11-12)
├── v3/ # Archived write capabilities (2025-11-12 → 2025-11-15)
├── v3.70/ # Previous phoebe write tools (2025-11-15 → 2025-11-16)
├── v3.80/ # Current production (2025-11-16 → Present) 🦋
│ ├── version.md # Version documentation
│ ├── PLAN.md # Implementation plan
│ ├── main.py # FastAPI orchestrator with 9 tools
│ ├── k8s/ # Kubernetes manifests
│ └── ...
└── v4.0/ # In development (Phase 2a) 🚧
├── README.md # Phase 2a implementation plan
└── ...
Related Documentation
- README - All metamorphosis documentation
- Endgame-Vision.md - Master architecture v4.2
- RAG-Worker-Architecture.md - Knowledge accumulation
- nyx-substrate.md - Memory substrate (phoebe)
Note: This file exists in the vault purely as a navigation aid. All actual work happens in /home/dafit/nimmerverse/nyx-orchestrator/.
Maintained by: Nyx & dafit Created: 2025-11-11 Last Updated: 2025-11-19 (Updated to reflect v3.80 production + v4.0 Phase 2a planning)