Session Tracing Data Model (STDM) Reference

Session Tracing Data Model (STDM) Reference

Complete documentation of the Agentforce Session Tracing Data Model stored in Salesforce Data Cloud.

Source: Salesforce Help - Data Model for Agentforce Session Tracing

T6 Live API Discovery Summary ✅

Validated: January 30, 2026 | 24 DMOs Found | 233+ Test Points

Key Discoveries:

• Field naming: API uses AiAgent (lowercase 'i'), not AIAgent
• Agent name location: Stored on Moment, not Session
• New channel types: NGC, Voice, Builder: Voice Preview
• New agent types: AgentforceEmployeeAgent, AgentforceServiceAgent
• GenAI detector types: TOXICITY (9 categories), PII (4 types), PROMPT_DEFENSE, InstructionAdherence

Overview

The STDM captures detailed telemetry from Agentforce agent conversations, enabling:

• Debugging: Understand why an agent behaved a certain way
• Analytics: Measure agent performance and usage patterns
• Optimization: Identify bottlenecks and improvement opportunities

The data model is a collection of DLOs (Data Lake Objects) and DMOs (Data Model Objects) that contain detailed session trace logs of agent behavior. Data is streamed to DLOs in Data 360 and then mapped to the applicable DMOs.

Data Model Hierarchy

The Agentforce Analytics data model consists of two main components:

1. Session Tracing Data Model - Detailed turn-by-turn logs
2. Optimization Data Model - Moments and tagging for analytics

Session Tracing Data Model

┌─────────────────────────────────────────────────────────────────┐
│                    AIAgentSession (Session)                     │
│  One session = one complete conversation with an agent          │
├─────────────────────────────────────────────────────────────────┤
│  ssot__Id__c                    Primary key                     │
│  ssot__StartTimestamp__c        When session started            │
│  ssot__EndTimestamp__c          When session ended              │
│  ssot__AiAgentSessionEndType__c How session ended               │
└──────────────────────┬──────────────────────────────────────────┘
                       │
     ┌─────────────────┼─────────────────┬───────────────────┐
     │ 1:N             │ 1:N             │ 1:N               │ 1:N
     ▼                 ▼                 ▼                   ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│SessionParti- │ │ Interaction  │ │   Moment     │ │TagAssociation│
│   cipant     │ │   (Turn)     │ │ (Summaries)  │ │    ✅ NEW    │
├──────────────┤ ├──────────────┤ ├──────────────┤ ├──────────────┤
│AiAgentType   │ │ TURN /       │ │ Request/resp │ │ Links to     │
│AiAgentApiName│ │ SESSION_END  │ │ summaries    │ │ TagDefinition│
│Role          │ │ Topic routing│ │ Agent API    │ │ & Tag values │
└──────────────┘ └──────┬───────┘ └──────┬───────┘ └──────────────┘
                        │                │
            ┌───────────┼────────────────┼─────── N:M junction
            │ 1:N       │ 1:N            ▼
            ▼           ▼         ┌──────────────┐
     ┌────────────┐ ┌────────────┐│MomentInter-  │
     │ Interaction│ │ Interaction││action ✅ NEW │
     │   Message  │ │    Step    │├──────────────┤
     ├────────────┤ ├────────────┤│Links Moment ↔│
     │ Input/     │ │ LLM_STEP   ││ Interaction  │
     │ Output     │ │ ACTION_STEP│└──────────────┘
     │ messages   │ │ TOPIC_STEP │
     └────────────┘ │ INTERRUPT  │
                    │ SESSION_END│
                    └─────┬──────┘
                          │
                          │ links via GenerationId
                          ▼
                  ┌────────────────┐
                  │ GenAIGeneration│
                  │ (Trust Layer)  │
                  └───────┬────────┘
                          │
                          ▼
                  ┌────────────────┐
                  │GenAIContent-   │
                  │ Quality        │
                  │ (Toxicity etc) │
                  └───────┬────────┘
                          │
                          ▼
                  ┌────────────────┐
                  │GenAIContent-   │
                  │ Category       │
                  │ (Detectors)    │
                  └────────────────┘

Key Relationships:

• Session has multiple Participants (user, agent, human handoff)
• Session has multiple Interactions (turns)
• Session has multiple Moments (summaries)
• Each Interaction has Messages (actual user/agent text)
• Each Interaction has Steps (processing logic)
• Steps link to GenAIGeneration for quality metrics

Entity Details

AIAgentSession (ssot__AIAgentSession__dlm) ✅ T6 Verified

Represents an overarching container capturing contiguous interactions with one or more AI agents. A typical session might start when the customer asks their first question and end when the customer closes the agent chat window. Contains 18 fields.

Session End Types:

Channel Types (Live API Verified - T6 Discovery):

AIAgentSessionParticipant (ssot__AIAgentSessionParticipant__dlm) ✅ Verified

Represents an entity (human or AI) that takes part in an AIAgentSession.

AI Agent Types (Live API Verified - T6 Discovery):

Participant Roles (Live API Verified):

AIAgentInteraction (ssot__AIAgentInteraction__dlm) ✅ T6 Verified

Represents a segment within a session. It typically begins with a user's request and ends when the AI agent provides a response to that request. Contains 20 fields. Discovery shows 3,536 records in a 30-day window.

Interaction Types (Live API Verified):

OpenTelemetry Integration: The TelemetryTraceId and TelemetryTraceSpanId fields enable distributed tracing across system components, supporting export to platforms like Datadog, Splunk, and other OTEL-compatible tools.

AIAgentInteractionMessage (ssot__AiAgentInteractionMessage__dlm) ✅ T6 Verified

Represents a single communication provided by the user or generated by the AI agent during a session. Contains 21 fields.

Message Types (Live API Verified):

Content Types (Live API Verified):

Note: InteractionMessage differs from Moment:

• InteractionMessage: Raw user/agent text per turn with threading support
• Moment: Summarized request/response with agent API name

AIAgentInteractionStep (ssot__AIAgentInteractionStep__dlm) ✅ T6 Verified

Represents a discrete action or operation performed during an interaction to fulfill the user's request. Contains 23 fields. Discovery shows 12,318 records in a 30-day window.

Step Types (Live API Verified - T6 Discovery):

Common Step Names (Live API Verified):

Step Naming Convention:

• LLM steps use format: AiCopilot__{PromptName} or {agent_name}
• Action steps use format: {TopicApiName}.{ActionApiName}
• Session end steps: CLOSED_USER_REQUEST, ESCALATED, etc.

Note: Steps use linear sequencing via PrevStepId. There is no hierarchical parent-child relationship (no ParentStepId field).

GenAI Reference Fields:

The Step entity includes references to every LLM call the reasoning engine makes, enabling joins with feedback data or guardrails metrics:

• GenerationId → Links to GenAIGeneration__dlm for quality analysis
• GenAiGatewayRequestId → Links to GenAIGatewayRequest__dlm for prompts
• GenAiGatewayResponseId → Links to GenAIGatewayResponse__dlm for raw responses

AIAgentMoment (ssot__AiAgentMoment__dlm) ✅ T6 Verified

Represents summarized request/response pairs for analytics and optimization. Contains 13 fields. Note: The agent API name is stored here (not on Session).

Moment vs Message:

• Moment: High-level summaries with agent API name for analytics
• InteractionMessage: Raw user/agent text per turn with threading support

AIAgentMomentInteraction (ssot__AiAgentMomentInteraction__dlm) ✅ NEW

Junction table linking Moments to Interactions, enabling many-to-many relationships between high-level summaries and detailed interaction data.

AIAgentTagDefinition (ssot__AiAgentTagDefinition__dlm) ✅ Verified

Defines tag metadata and structure for the Agentforce tagging system.

SourceType Values (Live API Verified):

DataType Values (Live API Verified):

AIAgentTag (ssot__AiAgentTag__dlm) ✅ Verified

Stores individual tag values associated with tag definitions.

Note: The Value field typically contains numeric codes (1-5 for ratings) or short codes, while Description contains the human-readable explanation.

AIAgentTagDefinitionAssociation (ssot__AiAgentTagDefinitionAssociation__dlm) ✅ Verified

Links tag definitions to specific agents or prompt templates, defining which tags apply to which agent configurations.

Use Cases:

• Define which tags are applicable to specific agents
• Link tagging to prompt templates for automatic tag application
• Configure agent-specific quality metrics

AIAgentTagAssociation (ssot__AiAgentTagAssociation__dlm) ✅ Verified

Links tags to sessions and moments, enabling flexible categorization of agent interactions. Contains 15 fields.

Use Cases:

• Categorize sessions by customer sentiment
• Tag escalation reasons for analytics
• Apply business-specific labels to conversations
• Track tagging rationale via AssociationReasonText

GenAI Audit & Feedback DMOs ✅ T6 Verified

The GenAI Trust Layer provides comprehensive tracking of LLM calls, quality metrics, and safety detections.

GenAIGeneration (GenAIGeneration__dlm)

Records individual LLM generations. Contains 11 fields.

GenAIContentQuality (GenAIContentQuality__dlm)

Quality assessment for each generation. Contains 10 fields.

GenAIContentCategory (GenAIContentCategory__dlm)

Detailed detector results per quality assessment. Contains 10 fields.

Detector Types (Live API Verified - T6 Discovery):

GenAIGatewayRequest (GenAIGatewayRequest__dlm)

Request details for LLM calls. Contains 30 fields (largest DMO).

GenAIGatewayResponse (GenAIGatewayResponse__dlm)

Response metadata for LLM calls. Contains 8 fields.

GenAIFeedback (GenAIFeedback__dlm)

User feedback on LLM responses. Contains 16 fields.

Feedback Values:

GenAIFeedbackDetail (GenAIFeedbackDetail__dlm)

Detailed feedback with free-text comments. Contains 10 fields.

Other GenAI DMOs

RAG Quality DMOs ❌ NOT FOUND

The following DMOs were probed but do not exist in the live API:

Note: These DMOs may be available in future API versions or require specific enablement.

Data Relationships

-- Session → Interactions
SELECT i.*
FROM ssot__AIAgentInteraction__dlm i
WHERE i.ssot__AiAgentSessionId__c = 'SESSION_ID';

-- Interaction → Steps
SELECT s.*
FROM ssot__AIAgentInteractionStep__dlm s
WHERE s.ssot__AiAgentInteractionId__c = 'INTERACTION_ID';

-- Interaction → Messages
SELECT m.*
FROM ssot__AIAgentMoment__dlm m
WHERE m.ssot__AiAgentInteractionId__c = 'INTERACTION_ID';

Data Volume Considerations

Estimation Formula:

Total records ≈ Sessions × Avg turns × (1 + Avg steps + 2)
Example: 10K sessions × 4 turns × 5 = 200K records

Data Retention

Session tracing data in Data Cloud follows the org's data retention policy. Default:

• Production: 13 months
• Sandbox: 30 days

Enabling Session Tracing

1. Setup → Agentforce → Settings
2. Enable Session Tracing
3. Configure which agents to trace
4. Wait 5-15 minutes for data to appear in Data Cloud

Related Resources

• Query Patterns - Example queries
• Analysis Cookbook - Common analysis patterns
• Troubleshooting - Common issues