Agent Definition

Custom object AgentDefinition__c. Editable core configuration used to execute an AI agent.

Field	API name	Type	Description
Agent Execution Identity	ExecutionIdentity__c	Lookup	Decides who the agent’s runs act as. Set: the identity’s execution user, through its Named Credential. Blank: whoever starts the run. An identity in use cannot be deleted; deactivate it instead.
Allow Multiple Tool Calls Per Turn	AllowMultipleToolCalls__c	Checkbox	Whether one model turn may request several tool calls. Core runs them one after another as a batch and returns all results to the model together.
Developer Name	DeveloperName__c	Text	Stable API identifier for this agent definition.
Instructions	Instructions__c	LongTextArea	System instructions supplied to the model for every run.
Is Active	IsActive__c	Checkbox	Controls whether new runs can use this definition.
Max Model Calls	MaxModelCalls__c	Number	Maximum model invocations allowed for one run, or for each reply of a conversational agent.
Max Model Request Characters	MaxModelRequestCharacters__c	Number	Maximum characters Core assembles into one model request, including instructions, inputs, context, and transcript.
Max Retries	MaxRetries__c	Number	Optional maximum retry attempts for one retryable command executed by this agent.
Max Tool Calls	MaxToolCalls__c	Number	Maximum tool invocations allowed for one run, or for each reply of a conversational agent.
Max Total Tokens	MaxTotalTokens__c	Number	Maximum model tokens (input plus output, as the provider reports them) for one run, or for each reply of a conversational agent.
Model	Model__c	Text	API name of the active Agent Model record this agent calls. The record supplies the model adapter, Named Credential, provider model identifier, and prices.
Model Timeout Seconds	ModelTimeoutSeconds__c	Number	Callout timeout passed to the model provider for each model request.
Output Schema	OutputSchema__c	LongTextArea	JSON Schema required for the final model output.
Tool Authorization Policy Class	ToolAuthorizationPolicyClass__c	Text	Optional query-only class implementing IToolAuthorizationPolicy.
Transcript Turn Limit	TranscriptTurnLimit__c	Number	Maximum prior turns (tool exchanges, human responses, operator guidance) replayed to the model.
Wait for Reply (Hours)	ReplyWindowHours__c	Number	Makes the agent conversational. After each answer the run waits this many hours for the person’s next message, with the conversation so far in the transcript. Blank ends the run with its first answer.