dafit
7688ded93f
- Endgame-Vision.md: hardware timeline sync (RTX 6000 Dec 31), GPU naming standardized to PRO 6000 Blackwell, Memory-Gradient.md rename - nyx-metamorphosis/README.md: v6.0 refs, clean index, archived projects marked - nyx-metamorphosis/Nyx_Traits.md v2.0: aligned to v6.0, LoRA mapping, mythological children preserved - nyx-metamorphosis/RAG-Worker-Architecture.md: marked ARCHIVED, points to Memory-Gradient - nyx-metamorphosis/nyx-orchestrator.md: moved to deep archive 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
38 lines
1.1 KiB
Markdown
38 lines
1.1 KiB
Markdown
# RAG Worker Architecture
|
|
|
|
**Status**: 📦 ARCHIVED
|
|
**Superseded by**: [Memory-Gradient.md](../operations/Memory-Gradient.md)
|
|
|
|
---
|
|
|
|
## Historical Context
|
|
|
|
This was a pointer file to `/home/dafit/nimmerverse/rag-worker/` which contained the Phase 2a RAG accumulation architecture.
|
|
|
|
**What it was:**
|
|
- ChromaDB vector storage for decision trails
|
|
- Multi-organ decision pattern storage
|
|
- Substrate for LoRA training data
|
|
|
|
**Why archived:**
|
|
- Architecture evolved from multi-organ (v4.2) to single-model + LoRA (v6.0)
|
|
- RAG approach superseded by Memory-Gradient internalization
|
|
- Fresh implementation will follow new architecture
|
|
|
|
---
|
|
|
|
## Future Direction
|
|
|
|
The Memory-Gradient approach in v6.0 handles knowledge differently:
|
|
- RAG as temporary scaffold, not permanent architecture
|
|
- Internalization into LoRA weights over time
|
|
- Metacognitive routing decides RAG vs direct inference
|
|
|
|
See: [Memory-Gradient.md](../operations/Memory-Gradient.md) for current approach.
|
|
|
|
---
|
|
|
|
**Archived**: 2025-12-29
|
|
**Original**: 2025-11-10
|
|
**Reason**: Paradigm shift from multi-organ RAG to single-model Memory-Gradient
|