Understand Agentforce Observability KPIs and Tools
Observability isn't a single tool. It's a layered toolchain that includes infrastructure that captures agent activity, development tools to test your agent pre-deployment, and monitoring tools that surface trends and session-level details after deployment.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type. |
Knowing which layer does what is what keeps your agent observable from build through production. This article maps the full toolchain, and each section links to more detail.
Infrastructure
Before any observability tool can show or process data, they need the underlying agent telemetry infrastructure. Enable these tools so that the rest of the toolchain has data to work with.
| Tool | Where it appears | Description |
|---|---|---|
| Agentforce Session Tracing | Setup → Agentforce → Agentforce Session Tracing | Captures every turn, reasoning step, action, and message for every agent session. This must be enabled before any other observability tool has data. |
| Agent Platform Tracing | Setup → Quick Find → Einstein Generative AI → Einstein Audit, Analytics, and Monitoring Setup → Agent Platform Tracing toggle | Captures service-level telemetry across Atlas reasoning engine, Apex, Flows, Prompt Builder, Invocable Actions, AI Gateway, LLM Gateway, and DC Query Federator. |
| Audit and Feedback | Setup → Quick Find → Einstein Audit → Einstein Audit, Analytics, and Monitoring Setup | Captures LLM inputs/outputs, retrieval quality data, trust evaluations, guardrail outputs, and user feedback signals. Powers the Quality Scores and Trust in Agent Analytics. |
| Knowledge/RAG Quality Data and Metrics | Setup → Quick Find → Einstein Audit → Einstein Audit, Analytics, and Monitoring Setup → Knowledge/RAG Quality Data and Metrics | Scores retrieval quality like Answer Faithfulness, Answer Relevance, Context Relevance, processed via the Audit and Feedback pipeline. Powers the retrieval metrics in the Quality Scores. |
| Trust Layer | Setup → Quick Find → Einstein → Einstein Trust Layer | Scores instruction adherence and toxicity derived from LLM Gateway outputs, processed via the Audit and Feedback pipeline. Powers the Trust metrics in Agent Analytics. |
Drill down into each of these tools in Agentforce Observability Infrastructure.
Build & Test
Build & Test tools let you observe agent behavior during development, before production traffic exists.
| Tool | Where it appears | Description |
|---|---|---|
| Agentforce Builder | Agentforce Studio → Builder | The primary workspace for iterative agent development and testing. Run conversations in Simulate mode without writing data, or Live Test mode against real data end-to-end. |
| Interaction Summary | Agentforce Studio → Builder → Preview panel | Shows subagent selection, action execution, and reasoning for a test conversation. Drill into any item for step-by-step timing and subagent transitions. |
| Testing Center | Agentforce Studio → Agentforce Testing Center | Batch evaluation of knowledge retrieval, subagent and action selection, and task resolution, before deployment. Generates and scores utterances in a test suite using default, quality, and custom scorers. Custom scorers surface post-deployment in Agent Analytics. |
Drill down into each of these tools in Build and Test Observability Tools.
Observe
Once your agent is deployed, the Observe phase tools track trends across all sessions, so you can investigate specific conversations and drill into step-level detail.
| Tool | Where it appears | Description |
|---|---|---|
| Agent Analytics | Agentforce Studio → Observe & Optimize → Analytics | Aggregates performance data across all sessions and surfaces trends by outcome area in Tableau Next dashboards. Tabs vary by agent type and cover effectiveness, usage, quality, health, trust, and voice. |
| Agent Optimization | Agentforce Studio → Observe & Optimize → Optimization | Individual session view with filtering by intent, outcome, and quality score. Processed Sessions include intent clustering and quality scores; Unprocessed Sessions are available immediately. |
| Interaction Summary | Agentforce Studio → Observe & Optimize → Optimization session detail | Structured breakdown of a selected intent showing subagents triggered, actions triggered, quality score, and LLM-generated quality score reasoning. |
| Trace | Agentforce Studio → Observe & Optimize → Optimization session detail | Time-ordered record of every processing step with per-step timing and error text. Available immediately, including for sessions still being processed. |
| Agent Health | Agentforce Studio → Alerts | Create alerts and set thresholds so that you’re notified when agent activity breaches acceptable levels. |
| Monitor Agent Guardrails | Agentforce Studio → Observe & Optimize | Dedicated view for instruction adherence and task resolution trends, separate from the broader Analytics view. |
| Monitor Your Voice Enabled Agent | Agentforce tab in Omni Supervisor (Service Cloud console) | Near real-time supervisor view of active voice calls with live transcripts. |
| Data 360 Query Editor | Data 360 → Queries | Direct SQL access to all four underlying data layers for custom aggregations, cross-object joins, and analysis that the product UI doesn't surface. |
Drill down into each of these tools in Observe Phase Observability Tools.
The toolchain works best in sequence. Start with Infrastructure to ensure data is flowing, use Build & Test to catch issues before deployment, and use Observe to track performance and investigate after deployment.
