Configuration

Configuration Guide

Complete reference for all agent and capability settings.

Use This Page As Your Admin Baseline

Start with agent + provider settings, then tune safety and performance as usage grows.

Agent Settings

Identity, memory, behavior, and runtime controls.

Trust & Safety

PII masking and prompt safety thresholds.

Capability Design

Implementation type, schemas, and execution rules.

Admin Security Critical Reference

1. Configure core agent and LLM settings.
2. Enable trust and safety controls.
3. Define capability contracts and execution policies.
4. Validate in sandbox before production rollout.

Troubleshooting Guide Security Reference

Agent Configuration

Basic Settings

Field	Type	Description
Name	Text	Display name for the agent
DeveloperName__c	Text	Unique API identifier (no spaces)
AgentType__c	Picklist	Conversational, Function, Workflow, or Email
IsActive__c	Checkbox	Enable/disable the agent

AI Provider Settings

Field	Type	Description
LLMConfiguration__c	Lookup	Which AI provider configuration to use
MemoryStrategy__c	Picklist	How to manage conversation history
HistoryTurnLimit__c	Number	Number of conversation turns to remember

Memory Strategies

Strategy	Description	Best For
BufferWindow	Keeps last N turns verbatim	Short conversations, precise context
SummaryBuffer	Summarizes older turns	Long conversations, cost optimization

Behavior Settings

Field	Type	Description
IdentityPrompt__c	Long Text	Defines who the agent is (persona)
InstructionsPrompt__c	Long Text	How the agent should behave
EnableActionTransparency__c	Checkbox	Show tool execution details to users
EnableToolReasoning__c	Checkbox	Require LLM to explain tool selection for better transparency
AuditLevel__c	Picklist	None, Standard, or Detailed logging
EnableDependencyValidation__c	Checkbox	Enforce tool dependency graph at runtime
ToolDependencyGraph__c	Long Text	JSON dependency graph (approved)
EnableNextStepSuggestion__c	Checkbox	Injects _nextStepSuggestion into tools (experimental)

Performance Settings

Field	Type	Description
AsyncDispatchType__c	Picklist	High (Platform Events) or Low (Queueables)
EnableParallelToolCalling__c	Checkbox	Execute multiple tools simultaneously
MaxProcessingCycles__c	Number	Max LLM cycles per execution

Service User Context (Optional)

Field	Type	Description
RequiresServiceUserContext__c	Checkbox	Route execution through service user context via REST callout
ServiceUserNamedCredential__c	Text	Named Credential for loopback callouts

Trust & Safety

Field	Type	Description
PIIMaskingMode__c	Picklist	Hybrid, Schema-Only, or Pattern-Only
SensitiveClassifications__c	Multi-Select	Data classifications to mask
PIIPatternCategories__c	Multi-Select	Regex pattern categories to enable
PromptSafetyMode__c	Picklist	Block, Sanitize, Flag, or LogOnly
SafetyThreshold__c	Number	Threat score threshold (0.0–1.0)
SafetyPatternCategories__c	Multi-Select	Which jailbreak categories to enable

---

Async Dispatch Types

High Concurrency (Platform Events)

Best for production environments with many concurrent users.

• Handles thousands of simultaneous conversations
• Event-driven architecture
• Better scalability
• Harder to debug

Low Concurrency (Queueables)

Best for development, testing, and debugging.

• Sequential processing
• Full debug log support
• Guaranteed execution order
• Limited concurrent executions

---

LLM Configuration

Field	Type	Description
DeveloperName__c	Text	Unique identifier
NamedCredential__c	Text	Named Credential API name
ProviderAdapterClass__c	Text	Apex class for provider integration
DefaultModelIdentifier__c	Text	Model identifier (e.g., gpt-4o)
DefaultTemperature__c	Number	Creativity level (0.0 - 1.0)
IsActive__c	Checkbox	Enable this configuration

Temperature Guide

Value	Behavior	Use Case
0.0 - 0.3	Deterministic, focused	Data retrieval, classification
0.4 - 0.7	Balanced	General assistance
0.8 - 1.0	Creative, varied	Content generation

---

Capability Configuration

Basic Info

Field	Type	Description
CapabilityName__c	Text	Tool name shown to AI (use snake_case)
Description__c	Long Text	When and how to use this tool
ImplementationType__c	Picklist	Standard, Apex, or Flow

Implementation Types

Type	Description	Use Case
Standard	Built-in actions	Common Salesforce operations
Apex	Custom Apex class	Complex business logic
Flow	Salesforce Flow	No-code automation

Execution Settings

Field	Type	Description
HITLMode__c	Picklist	Human-in-the-Loop mode: blank (no HITL), Confirmation (LLM asks in chat), Approval (formal approval process), or ConfirmationThenApproval (both)
HITLNotificationPreference__c	Picklist	Controls when to send notifications for HITL actions: “Always Notify” (default) sends notifications for approvals, rejections, and errors; “Notify on Rejection Only” only sends notifications when actions are rejected. Only applies when HITLMode__c is “Approval” or “ConfirmationThenApproval”.
RunAsynchronously__c	Checkbox	Execute in separate transaction
FailFastOnError__c	Checkbox	Stop immediately on error
ExposureLevel__c	Picklist	External (visible to LLM), Internal (framework only), or Disabled

Configuration Fields

Field	Type	Description
BackendConfiguration__c	Long Text	Admin settings (JSON)
Parameters__c	Long Text	Tool parameters (JSON Schema)

---

Writing Effective Descriptions

The capability description is crucial for the AI to understand when to use a tool.

Good Description Example

Search for contacts in Salesforce by name, email, or account.
Use this capability when:
- User asks to find a contact
- User wants contact information
- User mentions a person's name in a business context

Do NOT use when:
- User is asking about accounts (use search_accounts instead)
- User wants to create a new contact (use create_contact instead)

Bad Description Example

Gets contacts

---

JSON Schema for Parameters

Parameters use JSON Schema format.

Basic Example

{
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "The contact's full name"
    }
  }
}

With Required Fields

{
  "type": "object",
  "required": ["email"],
  "properties": {
    "email": {
      "type": "string",
      "description": "Email address (required)"
    },
    "name": {
      "type": "string",
      "description": "Optional name filter"
    }
  }
}

With Enums

{
  "type": "object",
  "properties": {
    "priority": {
      "type": "string",
      "enum": ["High", "Normal", "Low"],
      "description": "Task priority level"
    }
  }
}

Complex Example

{
  "type": "object",
  "required": ["objectType"],
  "properties": {
    "objectType": {
      "type": "string",
      "enum": ["Account", "Contact", "Opportunity"],
      "description": "Salesforce object to search"
    },
    "filters": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "field": {"type": "string"},
          "operator": {"type": "string", "enum": ["=", "!=", "LIKE"]},
          "value": {"type": "string"}
        }
      },
      "description": "Search filters to apply"
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 100,
      "default": 10,
      "description": "Maximum records to return"
    }
  }
}

---

Backend Configuration Examples

GetRecords Action

{
  "objectApiName": "Contact",
  "defaultFields": ["Id", "Name", "Email", "Phone", "Account.Name"],
  "defaultLimit": 25
}

CreateRecord Action

{
  "objectApiName": "Task",
  "defaultValues": {
    "OwnerId": "{!$User.Id}",
    "Status": "Not Started"
  }
}

Flow Action

For Flow implementation, set ImplementationType__c to Flow and put the Flow API name in ImplementationDetail__c.

{
  "defaultInputValues": {
    "source": "AI_Agent"
  }
}