Local AI control layer for my home lab.
lux-core defines how my personal AI system behaves across:
- The LLM engine on the RTX 3090 (
home-llm-gateway-3090) - My SOC / cloud / networking labs
- My language and quant tools
It is the policy + orchestration layer, not the model weights.
Previously: SARAH-ANN Core.
Now renamed and aligned with my on-prem server identity.
- Run powerful models locally with strict control.
- Define what the AI is allowed to do and how it talks.
- Standardize how apps in my lab call the LLM gateway:
- SOC runbook assistant
- Spanish tutor
- Market deviation tools
- Future apps
Lux does not care which exact model is loaded; it cares about:
- Policies
- Tools
- Workflows
lux-core/
├── README.md
├── policies/
│ ├── alignment-policy.md # what the AI is and is NOT for
│ ├── privacy-policy.md # data handling, logging, retention
│ ├── safety-policy.md # refusal rules, harmful content, boundaries
│ └── autonomy-policy.md # what it can trigger or change in the lab
├── prompts/
│ ├── system-prompts.md # core system prompts (general / dev / soc / spanish)
│ ├── style-guidelines.md # tone, format, how responses should look
│ └── tool-directives.md # how the model should request tools (search, RAG, etc.)
├── tools/
│ ├── tool-catalog.md # list of tools lux is allowed to call
│ ├── tool-profiles/
│ │ ├── web-search.md
│ │ ├── rag-docs.md
│ │ ├── soc-lab-tools.md
│ │ └── quant-tools.md
│ └── access-rules.md # which app can use which tools
├── flows/
│ ├── overview.md # how requests flow through lux → gateway → tools
│ ├── soc-runbook-flow.md # ai-soc-runbook-qa interaction pattern
│ ├── spanish-tutor-flow.md # es-ai-immersion-lab interaction pattern
│ ├── market-deviation-flow.md # market deviation assistant pattern
│ └── generic-chat-flow.md # default “personal assistant” chat pipeline
├── gateway-integration/
│ ├── models-policy.md # which models are allowed (sizes, families, licenses)
│ ├── home-llm-gateway-3090.md # how lux talks to the gateway API
│ └── rate-limits.md # per-app and per-user limits
├── logs-spec/
│ ├── logging-policy.md # what can be logged, what must NOT be logged
│ ├── redaction-rules.md # how to scrub sensitive data before storage
│ └── review-process.md # how I review logs for tuning without leaking data
└── roadmap/
├── 2025-roadmap.md # concrete features and milestones
└── ideas.md # future experiments / stretch goals