Feature
Human-in-Loop
Route high-stakes AI decisions to human reviewers. Configure escalation paths, approval workflows, and notification channels for when AI alone isn't enough.
When AI Needs Human Backup
High-Value Transactions
"AI recommends trades > $100k"
→ Require manager approval
Sensitive Topics
"Medical or legal advice detected"
→ Route to compliance review
Edge Cases
"Low confidence score from AI"
→ Queue for human verification
Customer Escalation
"User requests human support"
→ Seamless handoff with context
Flexible Escalation Rules
policy.yaml
policies:
- name: high-risk-escalation
action: ESCALATE
conditions:
- type: keywords
values: [medical_advice, legal_advice, financial_recommendation]
- type: confidence_below
threshold: 0.7
escalation:
# Who to notify
notify:
- channel: slack
webhook: $SLACK_COMPLIANCE_WEBHOOK
- channel: email
to: compliance@company.com
# Approval workflow
require_approval: true
approvers:
- role: compliance_officer
- role: team_lead
# Timeouts
timeout_minutes: 30
timeout_action: BLOCK # or ALLOW with warning
# Include context
include_context:
- original_input
- ai_response
- policy_triggered
- risk_scoreNotification Channels
Slack
Real-time alerts to channels or DMs
Detailed notifications with full context
Microsoft Teams
Integrate with your M365 workflow
PagerDuty
Critical alerts with on-call routing
Webhook
Custom integration with any system
Tork Dashboard
Built-in review queue and approval UI
Approval Workflow
Policy Triggered
AI output matches escalation criteria
Request Queued
Held for review, user notified of delay
Reviewers Notified
Alert sent via configured channels
Human Decision
Approve, modify, or reject the AI response
Response Delivered
Approved content sent to user with audit trail
Add Human Oversight
Keep humans in control of high-stakes AI decisions.