Skip to main content

Scenarios

Eight conversations showing how the agent routes, responds, and adapts turn by turn. Each turn shows the actual chat, the graph routing path, and why the runtime or specialist agent made its decision.

All response-style names match the shipped TherapeuticAgent styles. Routing traces show the actual runtime stage sequence the turn would follow through the pipeline — see Execution Flows for what runs inside each stage. For the full catalog of 13 guided exercises with steps and trigger keywords, see Guided Exercises.

Routing traces are illustrative

The routing traces below are written for clarity, not strict parity with what the TherapeuticAgent chooses on a given turn. The crisis gate, the exercise step-state classifier, and the active-exercise fast path are all deterministic — those traces are accurate. The specific style picks (supportive vs. reflective vs. psychoeducation) are LLM-primary, so the live agent could land on a slightly different style for the same input. The annotations explain the agent's decision rule, not a guarantee of which response style fires every time.

therapeuticcrisis