Agent Decision Explainability

The Agent Decisions page answers the question: "Why did my agent choose that tool / take that action?"

For each decision event in a session, Prysm shows:

  1. Context — the preceding events (LLM calls, tool results) that led to this decision
  2. The Decision — what the agent chose to do
  3. Consequence — what happened after the decision
  4. Triggering LLM Call — the specific LLM call that produced this decision

How It Works

Decision events are recorded when:

  • The agent selects a tool (tool_call events)
  • The agent makes an explicit decision (decision events)
  • The agent delegates to a sub-agent (delegation events)

For each decision, Prysm looks backward through the event sequence to find the closest preceding LLM call — this is the model invocation that likely produced the decision.

Dashboard

Navigate to Dashboard → Decisions, select a session, and click any decision card to expand the full context view.

← Tool PerformanceWorkflow Graph →