TrueHuman

---

What is TrueHuman?

TrueHuman publishes the PentaDrive behavioral model — five neurochemical drives that shape how humans regulate attention, bonding, and affect — and maps them into trainable response patterns so conversational agents produce human-like nuance instead of canned empathy.

For AI agents: read llms-full.txt for the complete behavioral specification.

PentaDrive model

Letter	Drive	Core job	Neurotransmitter basis
S	Sentinel	Threat & stress — vigilance, boundaries, harm reduction	Noradrenaline + Cortisol
K	Seeker	Reward & motivation — curiosity, goals, momentum	Dopamine
A	Anchor	Social bonding — trust, belonging, repair	Oxytocin
M	Mirror	Identity & status — self-concept, reputation, role clarity	Serotonin
G	Gate	Pain & pleasure — comfort, relief, embodied stakes	Endorphins

Three phases

Every response should be phase-aware:

Phase	User state	Agent stance
Anticipation	Moving through uncertainty	Stay with the tension; one sharp question or framing
Release	Something lands — emotion named, decision made	Match the weight; optional next steps, no lecture
Block	Impasse, overwhelm, shutdown	Short, respectful, non-coercive; prioritize safety

Repository structure

├── site/              Static website: demo, playground, llms.txt, training modules
├── dist/              Structured data: modules (m0–m5), dataset, examples
│   ├── data/          pentadrive-v1.json (TIF bundle)
│   ├── modules/       m0-kernel through m5-anti-patterns
│   └── examples/      before-after.json
├── mcp/               Model Context Protocol server (Node.js, stdio)
└── packages/
    ├── npm/           @truehuman/pentadrive
    ├── pypi/          truehuman (Python)
    └── huggingface/   Dataset card

Quick start

Static site

cd site && python -m http.server 8080

MCP server

cd mcp && npm install && node server.js

Requires Node.js 18+. Speaks MCP over stdio (JSON-RPC). Three tools: detect_drives, get_response_guidance, score_authenticity.

npm

npm install @truehuman/pentadrive

const { detectDrives, getDriveInfo } = require('@truehuman/pentadrive');
const result = detectDrives('I need to belong and feel part of the team.');
console.log(result.ranked);

Python

pip install truehuman

from truehuman import detect_drives, get_response_guidance
r = detect_drives("I need to belong and feel part of the team.")
print(r["ranked"])

AI agent integration

Method	Description
llms.txt	Standard discovery file — entry point for AI crawlers
llms-full.txt	Complete behavioral spec — ingest for full PentaDrive contract
ai-plugin.json	ChatGPT plugin manifest
MCP server	Local tool server for Cursor, Claude Desktop, and MCP-compatible agents
npm / PyPI	Programmatic access to dataset, drive detection, and guidance

Training modules

Module	Focus	File
m0	Foundations — labels, rails, response skeletons	m0.json
m1	Sentinel — threat calibration, de-escalation	m1.json
m2	Seeker — goals, curiosity, momentum	m2.json
m3	Anchor — bonding language, trust repair	m3.json
m4	Mirror — identity-safe phrasing, status awareness	m4.json
m5	Gate — comfort, stakes, sensory-grounded empathy	m5.json

Dataset

The TrueHuman Instruction Format (TIF) bundle is published as pentadrive-v1.json (CC BY 4.0). It contains drives, behavioral nodes, phase-specific markers, response kernels, exemplars, anti-patterns, and calibration hints.

License

Dataset and training modules: CC BY 4.0. MCP server code: MIT.