Introduction to Tork
Tork is an AI governance platform that provides visibility, control, and compliance for AI agents at scale.
What is Tork?
Tork acts as a safety layer between AI agents and end users, providing comprehensive governance capabilities for autonomous AI systems. As AI agents become more capable and autonomous, organizations need robust tools to ensure they operate within safe and compliant boundaries.
PII Detection
Scan and redact 56 PII types across 13 countries and 3 industry profiles with regional profiles
Policy Enforcement
Define and enforce governance policies in real-time
Human Oversight
Human-in-the-loop approval workflows for high-risk actions
Cost Controls
Budget governance and spend tracking across providers
Memory Integrity
Trust scoring and drift detection for agent memory
Supply Chain
Dependency tracking and vulnerability scanning
Why Tork?
As AI agents become more autonomous, organizations face new challenges in managing risk, ensuring compliance, and maintaining trust. Tork addresses these challenges by providing:
Visibility
Know what your agents are doing with comprehensive audit logs, real-time monitoring, and detailed analytics.
Control
Set boundaries and enforce policies across all agents. Define what actions are allowed, require approvals for sensitive operations.
Compliance
Meet regulatory requirements with compliance receipts, detailed reporting, and a daily on-chain anchor that makes those records immutable.
Trust
Build confidence in AI systems through transparency, predictable behavior, and measurable governance metrics.
TORKING-X Metrics
Tork introduces TORKING-X, a standardized 9-metric system for measuring AI governance health. Each metric is scored 0-100 and weighted to produce an overall governance score.
| Metric | Weight | Description |
|---|---|---|
| TTamper-Evidence | 10% | Audit log integrity and chain verification |
| PPrivacy Protection | 15% | PII detection and redaction effectiveness |
| EEphemeral Identity | 10% | Session isolation and credential management |
| QQuality Guardrails | 15% | Output validation and safety checks |
| CCircuit Breakers | 15% | Failure handling and recovery mechanisms |
| SSupply Chain | 10% | Dependency security and verification |
| LTool Safety | 10% | Tool governance and permission controls |
| MMemory Integrity | 10% | Memory trust and drift detection |
| HHITL Enforcement | 5% | Human oversight and approval workflows |
TORKING-X Score
Quick Example
There are two ways to use Tork, and they are different packages. To govern on-device, install tork-governance and construct Tork. The decision is synchronous and, with no API key, makes no network calls at all:
from tork_governance import Tork
tork = Tork()
# The decision is made locally and returned synchronously
result = tork.govern("Contact john@example.com for details")
print(result.action) # 'allow', 'redact', or 'deny'
print(result.output) # "Contact [EMAIL] for details"To have Tork make the decision server-side and record a receipt, use the cloud SDK @torknetwork/sdk (npm install @torknetwork/sdk), whose entry class is TorkClient. There is no Python cloud SDK — from Python, call the REST API with a Bearer token.
import { TorkClient } from '@torknetwork/sdk';
const client = new TorkClient({ apiKey: process.env.TORK_API_KEY });
const result = await client.govern('Contact john@example.com for details', {
mode: 'redact',
});
console.log(result.action); // 'allow' | 'redact' | 'deny'
console.log(result.output); // 'Contact [EMAIL] for details'
console.log(result.receipt_id);