AI That Asks Before It Acts
Agent Hub is your mission control for AI. See every pending action, approve or reject with one click, and watch your AI get smarter from your feedback.
AI That Learns From You
Every approval or rejection teaches your AI. Over time, it learns your preferences—what messages to auto-approve, when to escalate, and how to handle edge cases just like you would.
How Agent Hub Works
AI Identifies an Action
Your AI decides to send a message, book an appointment, or take another action.
Action Enters Queue
Based on your autonomy settings, the action either auto-executes or waits for your approval.
You Review (Optional)
In Agent Hub, see the pending action with full context. Approve, reject, or edit.
AI Learns
Your decision is recorded. Over time, similar actions are handled automatically.
The Approval Queue
Pending actions appear in a queue with full context:
Customer asked about haircut prices. AI wants to send service menu.
New customer wants 2pm tomorrow. AI is booking with Mike.
Automated rebooking reminders for overdue customers.
Risk-Based Filtering
Each action is categorized by risk level so you can focus on what matters:
- •Answering routine questions
- •Sending appointment confirmations
- •Adding notes to customer profiles
- •Booking appointments
- •Updating customer details
- •Sending promotional messages
- •Cancelling bookings
- •Issuing refunds
- •Deleting customer data
- •Bulk operations
- •Modifying business settings
- •Actions affecting multiple customers
Autonomy levels (advanced)
Controls offers recommended modes. Use the advanced override to set these levels directly:
Observer
Read-only. Bizy suggests actions; approvals required for any change.
Assistant
Low-risk actions auto-execute. Medium and higher risk need approval.
Trusted
Low and medium risk auto-execute. High risk needs approval.
Autonomous
Most actions auto-execute. Critical actions still need approval.
Activity Timeline
See Everything
Every action your AI takes—approved, rejected, or auto-executed—is logged.
Time-Based View
Filter by last hour, day, week, or custom range.
Performance Stats
Track approval rates, accuracy, and actions per day.
Error Tracking
See failed actions and understand what went wrong.
Approval Actions
Approve
Execute the action as proposed. AI learns this is correct.
Reject
Don't execute. Optionally add a reason so AI learns why.
Edit & Approve
Modify the action before executing. AI learns the correction.
Evidence Viewer
For every AI action, you can see exactly what knowledge sources informed the decision.
Source Citations
See which knowledge base entries, documents, or FAQs the AI used to craft its response.
Confidence Breakdown
View the confidence score for each cited source and understand why the AI chose that information.
Source Freshness
Check when the cited sources were last updated to ensure responses are based on current information.
Intent-Based Escalation
Not all conversations are equal. The AI detects conversation intent and adjusts escalation behavior accordingly.
Low-Risk Intents
FAQ questions, service inquiries, availability checks
These can be handled at lower confidence thresholds.
Medium-Risk Intents
Booking requests, rescheduling, special requests
Standard approval thresholds apply.
High-Risk Intents
Refund requests, complaints, cancellations, billing issues
Auto-escalate for review regardless of confidence level.
How it works: The AI analyzes conversation context to detect intent, then applies the appropriate approval threshold automatically. A refund request will always require your review, even if the AI is confident in how to respond.
Pro Tip
Start at Level 1 (Observer) for the first week. As you approve actions, AI learns your patterns. After 50-100 approvals, consider moving to Level 2 (Assistant). Most businesses reach Level 3 (Trusted) within a month.