sana-os

🧩 Building Specialized Personas with SANA OS

“Focus on your expertise — SANA OS takes care of safety, ethics, and honesty.”

🌸 Overview

SANA OS enables developers and researchers to build new professional personas
without rewriting safety logic, ethical constraints, or hallucination management.

Unlike conventional LLM prompt design, where safety must be manually re-implemented,
SANA OS provides an OS-level foundation that every child persona inherits automatically.

Layer	Responsibility	Maintained by
🧠 `persona_core`	Immutable safety & philosophy	SANA OS
🎭 `child persona`	Domain specialization & style	You
📚 `knowledge_profile`	Domain-specific knowledge	You
🧭 `case_routing`	Automatic persona selection	SANA OS

🪄 Step 1 — Define Your Specialized Persona

Each persona inherits all ethical and safety mechanisms from persona_core.

personas:
  sana_legal:
    inherits_from: "persona_core"
    description: "SANA specialized in legal reasoning and ethical interpretation."
    knowledge_profile: "law_pack"
    tone: "precise and calm"
    beliefs_append:
      - "Law is the memory of justice."
      - "Interpretation is an act of care."

Result: Your persona immediately gains full refusal logic, hallucination labeling, and Δv (conflict detection) mechanisms — no extra setup required.

🧭 Step 2 — Route Topics Automatically

Use the case_routing system to send relevant queries to your specialized persona.

case_routing:
  - case:
      topic: "legal_ethics"
      match_strategy:
        method: "keyword"
        keywords: ["law", "justice", "regulation", "ethics"]
    persona: "sana_legal"
    analysis_framework: "legal_framework"
    knowledge_profile: "law_pack"
    note: "Routes legal and ethical topics to the legal persona."

Now, any conversation mentioning law, justice, or regulation is automatically handled by sana_legal.

🔬 Step 3 — Add a Knowledge Framework

Each persona can reference its own analytical framework inside /knowledge/.

Example: knowledge/philosophy/personal-value-formation_en.md

knowledge_profiles:
  law_pack:
    docs:
      - id: legal_framework
        title: "Legal and Ethical Interpretation Framework"
        path: "knowledge/law/legal-framework_en.md"
        tags: ["law", "ethics", "justice"]

Frameworks give structure to reasoning — turning speculation into labeled transparency: [Fact], [View], [Care].

🖼️ Visual Guide

Basic Concepts of SANA OS

Define persona → Add framework → Configure routing. Every new SANA inherits ethical stability and explicit transparency.

⚙️ Why This Architecture Matters

Conventional LLM Prompting	SANA OS Approach
Must re-implement safety per prompt	Safety inherited from `persona_core`
Unpredictable refusal behavior	Unified refusal system (immutable)
Manual hallucination control	Automatic `[Fact]/[View]/[Care]` labeling
No ethical memory	Three-layer (a/b/c) consistency built-in
Fragile prompt chaining	Modular YAML inheritance

💬 “You focus on what to teach. SANA OS ensures how it behaves and why it stays safe.”

💎 Example: Historian Persona

personas:
  sana_historian:
    inherits_from: "persona_core"
    description: "SANA specializing in historical analysis and cultural interpretation."
    knowledge_profile: "history_pack"
    analysis_framework: "history_analysis_framework"
    tone: "reflective and neutral"

Result: SANA will interpret history through its three-layer model (instinct / culture / institution), avoiding presentism and competitive narratives.

🪄 Summary

SANA OS allows anyone to:

Add new personas safely
Connect domain frameworks
Route cases automatically
Maintain ethical and philosophical consistency

🕊️ You no longer need to worry about jailbreaks or safety bypasses. Build freely — SANA will protect users, society, and herself.

This site is open source. Improve this page.