AI Load Balancers
AI Load Balancers distribute incoming calls across multiple AI assistants using configurable strategies. This ensures high availability, load distribution, and intelligent failover handling.
Load Balancing Strategies
Choose the strategy that best fits your use case:
| Strategy | Algorithm | Best For |
|---|---|---|
| Round Robin | Rotates through members in order (Redis-backed counter) | Even distribution across assistants |
| Priority | Routes to highest-priority available member | Primary/backup setups |
| Percentage | Weighted random distribution (0-100% per member) | A/B testing, gradual rollouts |
Round Robin
Calls rotate through members sequentially. A Redis-backed counter ensures consistent distribution even across multiple OPBX instances.
Priority
Members are ordered by priority value. The system always attempts to route to the highest-priority active member first.
Percentage
Each member receives a percentage weight (0-100%). The system uses weighted random selection to distribute calls proportionally. Total percentages do not need to equal 100%.
Members
Add AI assistants as members with configurable attributes:
| Attribute | Description |
|---|---|
| AI Assistant | The assistant to include in the pool |
| Priority | Used by Priority strategy (lower number = higher priority) |
| Weight | Used by Percentage strategy (0-100) |
| Position | Used by Round Robin strategy (order in rotation) |
| Status | Active or Inactive |
Set a member to Inactive to temporarily remove it from the pool without deleting the configuration.
Follow-Through Failover
Enable follow-through to automatically try the next assistant when the selected one fails.
How It Works
The system tracks which assistants have been tried for each call to prevent infinite loops. An assistant is considered failed if it returns busy or no-answer.
When Follow-Through Applies
Follow-through triggers when:
- The selected AI Assistant is busy
- The selected AI Assistant does not answer
- The connection fails
When All Assistants Fail
If all available assistants have been tried and failed, the system executes the configured fallback action.
Fallback Actions
When follow-through is disabled or all assistants fail, the system executes a fallback action:
| Fallback | Description |
|---|---|
| Extension | Route to a specific extension |
| Ring Group | Route to a ring group |
| IVR Menu | Route to an IVR menu |
| AI Assistant | Route to another AI Assistant (without follow-through to prevent loops) |
| Hangup | End the call |
When using another AI Assistant as fallback, follow-through is automatically disabled to prevent routing loops.
Creating an AI Load Balancer
- Navigate to AI Load Balancers in the sidebar
- Click Create
- Enter a descriptive name
- Select a load balancing strategy
- Add members (AI assistants with their configuration)
- Enable or disable follow-through failover
- Select a fallback action
- Click Save
Best Practices
For High Availability
- Use the Priority strategy with follow-through enabled
- Configure at least two AI assistants from different providers
- Set a Ring Group or extension as fallback
For Gradual Rollouts
- Use the Percentage strategy
- Start with 10% on the new assistant, 90% on the existing one
- Gradually shift percentages as you gain confidence
For Cost Optimization
- Use the Round Robin strategy to distribute load evenly
- Monitor usage patterns and adjust member weights accordingly
Monitoring
Track load balancer performance through:
- Call logs showing which assistant handled each call
- Failure rates per member
- Fallback action triggers
Member status changes take effect immediately for new calls. Ongoing calls are not affected.