rb-local-llm-persona-demo

© 2025–2026 Right Business Pte. Ltd. All rights reserved.

Reference implementation demonstrating persona enforcement for locally run large language models using a fixed Modelfile execution surface.

Status: Stable reference implementation. Not intended for production deployment.

Scope

• Local LLM only (bring your own runtime and model)
• No UI
• No hosting
• No APIs
• No compatibility guarantees

Intended Audience

• Engineers familiar with local LLM runtimes
• Developers evaluating persona enforcement architecture
• System designers exploring Modelfile-based control surfaces

Assumptions

You already know how to:

• Run a local LLM (e.g. Ollama, llama.cpp, LM Studio)
• Inspect and adapt Modelfiles
• Resolve model- or runtime-specific differences yourself

Purpose

The goal is to demonstrate how a fixed persona can be enforced at the Modelfile and behavioral-architecture level across permissive local LLM runtimes.

The canonical active Modelfile is:

• Modelfile-rouhe

All other Modelfiles in this repository are archived under /archive/modelfiles and are retained strictly for lineage and auditability. They are deprecated and must not be used for deployment.

This repository is not:

• A product
• A framework
• A supported chatbot
• A turnkey solution

Architecture

The authoritative behavioral specification for the persona implemented in this repository is defined in:

• docs/ROUHE_ARCHITECTURE.md

This document is platform-agnostic and supersedes earlier prompt-bound or GPT-specific persona artefacts, which are preserved under /archive/gpt-era for historical reference only.

License

All rights reserved.

See LICENSE and NOTICE for full legal terms.