AOIA-Core: local provenance/evidence-boundary runtime as a complementary pattern to NeMo Guardrails

[luciferprosun]

Hi NeMo Guardrails community,

I would like to share AOIA-Core for feedback as a complementary open-source pattern, not as a replacement for NeMo Guardrails.

Repo: https://github.com/luciferprosun/AOIA-Core

AOIA-Core is a local-first epistemic-control runtime for AI-assisted technical workflows. The current focus is on:

• deterministic Linux/RHCSA local retrieval,
• provenance-chain discipline,
• evidence vs reasoning boundary separation,
• contradiction tracking,
• human review surfaces,
• and a planned read-only RHCSA/Linux command grammar layer for detecting suspicious or hallucinated command shapes.

What AOIA-Core is not:

• not a replacement for NeMo Guardrails,
• not production-ready,
• not AGI,
• not a truth engine,
• not a shell execution framework,
• not a scientific validation platform.

Current public state:

• stable checkpoint: d7e3448
• safety tag: nlnet-safe-d7e3448
• tests: 146 OK, skipped=4

The project is being developed by a small independent builder on limited hardware, so the design is intentionally conservative, local-first, and narrow in scope. This post is not a funding request — it is a technical feedback request.

My question to the community:

Would there be interest in a short technical design note comparing NeMo Guardrails-style policy/constraint patterns with an audit-oriented local provenance and evidence-boundary runtime like AOIA-Core?

I would especially appreciate feedback on established patterns around:

• local provenance tracking,
• evidence/reasoning separation,
• contradiction tracking,
• deterministic retrieval vs model-output boundaries,
• read-only command-shape validation,
• and how to avoid overclaiming “truth” or “safety” in guardrail-like systems.

If useful, I can prepare a focused comparison note rather than opening an issue.

Thank you for any feedback.