Glossary
Agent
The caller persona that talks to the target — the simulated customer. Not to be confused with the agent you are testing (we usually call that the “target agent” in docs to avoid ambiguity). An agent has voice, responsiveness, ambient sound, and max-call-duration settings.
Scenario
One situation you want to test. Holds the description, agent assignment, target phone number (for voice), success criteria, and any custom functions. Lives in the Scenarios page in the portal.
Design style
How a scenario is authored. Two values:
- Graph — visual canvas of nodes and branches. Deterministic, scripted.
- Free-flow — natural-language description. Persona is generated at runtime.
Test path
One specific narrative through a scenario. A graph scenario can have many paths (one per branch combination). A free-flow scenario auto-generates several alternative paths from the description so you cover the space.
Test run
A single execution of a test path against the target agent. Produces a transcript, audio (voice only), tool-invocation log, latency stats, and a success / fail evaluation.
Report
The viewable artifact for one or more runs. Single-run reports show transcript + audio + grading. Aggregated reports trend pass rate, surface regressions, and roll up across paths.
Custom function
A caller-side tool, declared per free-flow scenario, that the simulated caller can invoke during the conversation. Example: getDate that fetches the current date from your API and lets the caller speak it during the call. See the full guide.
Built-in tool
Two tools every caller has automatically: end_call (hang up) and press_digit (DTMF for IVR navigation). Read-only — you can’t edit or remove them.
Response variable
A name → JSON path mapping on a custom function that exposes a value from the function’s response as {{name}} for use in the caller’s reply. The canonical mechanism for embedding tool results into spoken / written replies.
Vibe testing
A rapid exploratory mode where you provide a high-level intent in natural language and the platform generates scenarios on the fly using your own LLM key. See Vibe testing.
Outbound vs inbound
- Outbound scenarios place a call to the target agent.
- Inbound scenarios accept a call from the target’s IVR / system.
Workspace
Each user gets an isolated workspace with their own database. Nothing is shared across workspaces unless explicitly exported / imported.