Skip to main content

​
Configuration

Zild Assist configuration determines how your AI agents behave, how they interact with systems, and how conversations are routed operationally. This document covers:
  • Agent configuration
  • Knowledge management
  • Tool configuration
  • Escalation rules
  • Conversation routing
  • Workflow triggers
  • Operational controls
All configuration is Tenant-scoped and role-protected.

​
Agent Configuration

Agents are the intelligence core of Zild Assist. Navigate to: Assist → Agents → Select Agent Each Agent contains multiple configurable components.

​
1. System Instructions

System Instructions define:
  • Agent role (Support, SDR, Sales, Internal Assistant)
  • Tone and communication style
  • Behavioral constraints
  • Compliance requirements
  • Escalation conditions
Best practices:
  • Be explicit about allowed actions
  • Define what the agent must never do
  • Provide structured conversation goals
  • Include escalation triggers
Example goals:
  • Capture name, email, and company before qualifying
  • Confirm identity before sharing account data
  • Escalate billing disputes automatically
Clear instructions improve response consistency.

​
2. Knowledge Base Configuration

Knowledge improves factual grounding and reduces hallucinations. You can configure:
  • FAQ documents
  • Policy documents
  • Product manuals
  • Structured internal documentation
Best practices:
  • Keep knowledge structured and organized
  • Avoid conflicting information
  • Update documentation regularly
  • Separate public vs internal knowledge
Knowledge changes do not require channel reconfiguration.

​
3. Tool Configuration

Tools allow Agents to perform deterministic actions. Tools may include:
  • CRM creation
  • Database queries (SQL Server / PostgreSQL)
  • Order lookup
  • Contract generation
  • API calls
  • Internal system triggers
Each tool requires:
  • Secure credentials (stored per Tenant)
  • Defined input schema
  • Expected output structure
Best practices:
  • Use parameterized queries
  • Limit database permissions to read-only when possible
  • Validate required fields before tool execution
  • Log tool execution results
Tools bridge conversational AI with structured business logic.

​
Escalation Configuration

Escalation ensures controlled transition from AI to human. Navigate to: Assist → Agent → Escalation Rules Escalation can be triggered by:
  • User explicitly requesting human support
  • Sensitive keywords
  • Low confidence threshold
  • Tool execution failure
  • Policy constraints
Escalation routing options:
  • Assign to group
  • Assign to specific user
  • Notify Supervisor
  • Close with callback instruction
Conversation history remains intact during escalation.

​
Conversation Routing

Routing determines how conversations are distributed. Routing rules can include:
  • Channel-based routing
  • Group assignment
  • Skill-based routing (if configured)
  • Load balancing within group
Role-based visibility:
  • Administrator → All conversations
  • Supervisor → Group conversations
  • User → Assigned conversations only
Ensure routing aligns with operational structure.

​
Workflow Trigger Configuration

Workflows allow automation beyond conversational replies. Navigate to: Platform → Workflows Triggers can include:
  • Message received
  • Lead qualified
  • Tool executed
  • Conversation closed
  • Escalation triggered
Actions may include:
  • Send webhook
  • Update CRM
  • Send contract
  • Create task
  • Notify team
Workflows are deterministic and auditable.

​
Channel-Specific Configuration

While agent logic is channel-agnostic, certain behaviors may vary:

​
WhatsApp

  • Session window handling
  • Template messaging outside session
  • Media restrictions

​
Webchat

  • Domain restriction
  • Session persistence
  • UI customization
Ensure configuration respects provider constraints.

​
Integration Configuration

If using integrations, verify:
  • Azure OpenAI or OpenAI configured
  • Twilio credentials validated
  • Database credentials secured
  • ElevenLabs configured for voice (if applicable)
Integration failures should trigger:
  • Logged errors
  • Optional escalation
  • Retry logic (if configured)

​
Operational Controls

Zild Assist allows ongoing operational tuning. Supervisors and Administrators can:
  • Monitor conversations
  • Adjust escalation rules
  • Refine system instructions
  • Update knowledge
  • Optimize workflows
Configuration changes should follow internal governance policies.

​
Versioning & Change Management

For enterprise environments:
  • Document configuration changes
  • Test updates in sandbox before production
  • Monitor impact after deployment
  • Maintain audit logs for traceability
Configuration governance improves long-term reliability.

​
Testing Configuration Changes

After modifying configuration:
  1. Run controlled test conversations
  2. Validate tool execution
  3. Verify escalation logic
  4. Confirm workflow triggers
  5. Review logs for unexpected behavior
Testing prevents operational disruption.

​
Configuration Best Practices

For stable deployment:
  • Keep instructions clear and structured
  • Avoid overly complex prompts
  • Separate AI reasoning from workflow automation
  • Restrict tool permissions
  • Define explicit escalation paths
  • Monitor performance metrics regularly
Well-structured configuration leads to predictable outcomes.

​
Summary

Zild Assist configuration controls:
  • How the Agent thinks
  • What the Agent knows
  • What the Agent can execute
  • When the Agent escalates
  • How conversations are routed
  • How workflows automate business actions
Proper configuration transforms Zild Assist from a conversational tool into a structured, AI-driven operational system.