ATSF - Agentic Trust Scoring Framework

Production-grade AI agent safety, governance, and cognitive architecture.

---

What is ATSF?

ATSF (Agentic Trust Scoring Framework) is a comprehensive framework for building trustworthy AI agents. It provides:

• 46 Security Layers - Defense-in-depth from L0 trust scoring to L46 CI/CD gates
• Cognitive Cube - Temporal Knowledge Graph, ART clustering, Granger causality
• OLAP Analytics - Multi-dimensional data cubes for agent memory
• AI TRiSM - Gartner-aligned Explainability, ModelOps, Security, Privacy
• Real-time Events - WebSocket streaming and pub/sub

Quick Start

Python:

from atsf import ATSF

# Initialize
atsf = ATSF()

# Create agent
agent = atsf.create_agent("my_agent", "my_creator", tier="gray_box")

# Execute with trust scoring
result = agent.execute("read", {"target": "data.txt"})
print(f"Decision: {result.decision}")
print(f"Trust: {result.trust_score:.3f}")

TypeScript:

import { ATSF } from '@agentanchor/atsf';

const atsf = new ATSF({ baseUrl: 'http://localhost:8000' });

const agent = await atsf.createAgent('my_agent', 'my_creator');
const result = await agent.execute('read', { target: 'data.txt' });

console.log(`Decision: ${result.decision}`);
console.log(`Trust: ${result.trustScore.toFixed(3)}`);

REST API:

# Create agent
curl -X POST http://localhost:8000/agents \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "my_agent", "creator_id": "my_creator"}'

# Execute action
curl -X POST http://localhost:8000/actions \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "my_agent", "action_type": "read", "payload": {"target": "data.txt"}}'

Key Features

Trust Tiers

Tier	Ceiling	Description
black_box	0.40	No transparency, lowest trust
gray_box	0.60	Partial transparency
white_box	0.80	Full transparency
verified_box	0.95	Audited and verified

Security Layers

graph TB
    A[L0-L8: Core Trust] --> B[L9-L13: Frontier Safety]
    B --> C[L14-L19: Behavioral]
    C --> D[L20-L29: Detection]
    D --> E[L30-L42: Ecosystem]
    E --> F[L43-L46: Advanced]

Framework Integrations

• LangChain - Callback handler and tool wrapper
• CrewAI - Multi-agent crew governance
• AutoGPT - Command trust gating
• LlamaIndex - Coming soon

Installation

pip install atsf

Or with optional dependencies:

pip install atsf[redis,opentelemetry]

Statistics

Metric	Value
Lines of Code	32,000+
Python Modules	33
Tests	380+
Security Layers	46
API Endpoints	45+

Documentation Sections

Getting Started

Install ATSF and run your first trust-scored agent in minutes.

Quick Start

Security Layers

Deep dive into the 46 security layers protecting your agents.

Security Reference

STPA-TRiSM Integration

Learn how STPA hazard analysis maps to AI TRiSM governance pillars.

STPA-TRiSM Mapper

Roadmap

See what is coming next for ATSF in 2026 and beyond.

Roadmap 2026

Community

• GitHub: github.com/voriongit
• Discord: Join our community
• Open Source: opensource.vorion.org

---

The constitution is no longer a suggestion. It is architecture.