A user-side protocol for governable, inspectable, and repeatable interaction with Large Language Models.
- Version: v1.0
- Maturity: Stable — initial public normative release
- Classification: Interaction-layer protocol
- License: CC BY-SA 4.0 (see ATTRIBUTION.md)
- Maintenance: Active
- Audience: Professional, regulated, and authorship-sensitive use
This repository contains the authoritative reference documentation of the LLM Interaction Governance Protocol (LIGP).
LIGP is a formal interaction protocol that governs how a user interacts with a Large Language Model (LLM), with explicit separation between:
- persistent personal context (where enabled),
- session-local behavioral inference,
- explicit user instruction.
The protocol operates entirely at the human–LLM interaction layer and does not require access to model internals, training data, or system policies.
This repository is intended to function as standard infrastructure, not as editorial content, prompt collections, or a commercial product.
- Not prompt engineering
- Not a model alignment technique
- Not a memory optimization hack
- Not an AI safety or control system
- Not certification, compliance assurance, or legal advice
- Not a proprietary or closed framework
LIGP governs interaction behavior only.
- Inspection and governance of persistent personal context (where available)
- Visibility into session-local behavioral inference
- Explicit, stateless reconstruction of interaction behavior via instruction
- Deterministic interaction continuity without reliance on memory or profiling
- Model training or fine-tuning
- Dataset provenance
- Internal parameters or embeddings
- System-level safety policy enforcement
- Infrastructure security
LIGP consists of three orthogonal protocol components:
A governance mechanism to inspect, verify, correct, or delete any user-specific information treated as long-term personal context.
A reflective disclosure that makes session-local adaptation visible without implying persistence, profiling, or storage.
A reusable, explicit declaration of interaction constraints provided at session start, enabling deterministic behavior reconstruction across sessions.
These components are sequential, complementary, and non-interchangeable.
- Whitepaper
docs/whitepaper/ligp_whitepaper.md
Authoritative conceptual and architectural description of LIGP.
-
Regulatory Briefing Note
docs/regulatory/ligp_regulatory_briefing.md
Policy-facing overview for regulators, compliance leaders, and governance bodies. -
Enterprise Governance Mapping
docs/enterprise/ligp_enterprise_governance_mapping.md
Alignment of LIGP with ISO, EU AI Act, OECD, and enterprise governance frameworks.
- Licensing & Attribution
ATTRIBUTION.md
Conditions for reuse, attribution, and derivative work.
. ├── README.md
├── LICENSE
├── ATTRIBUTION.md
└── docs
├── whitepaper
│ └── ligp_whitepaper.md
├── regulatory
│ └── ligp_regulatory_briefing.md
├── enterprise
│ └── ligp_enterprise_governance_mapping.md
└── licensing
└── ligp_licensing_and_attribution.md
LIGP does not attempt to make AI systems fully transparent.
It makes interaction governable.
By separating storage, inference, and instruction, LIGP enables auditability, repeatability, and user-side control without requiring internal access to model implementations.
Any reuse or adaptation of LIGP must comply with the attribution and licensing terms defined in ATTRIBUTION.md.
The LICENSE file in this repository is the authoritative legal document governing use of LIGP.