Concepts

The concepts below apply to every Debriefer surface (MCP, API, GUI). Each section here is a one-paragraph primer; see the design handbook for the full treatment.

Agents

The interviewer persona. Configured with techniques, tone, and a model.

See: Designing the agent.

Blueprints

A reusable interview definition. Holds the graph (nodes + edges) and references an agent. Blueprints have a draft and published state.

See: Structuring the interview.

Graph

A DAG of question nodes connected by edges. Supports branching, probing, and multimodal input per node.

See: Structuring the interview.

Runs

A single instance of a published blueprint executing against a respondent. Has status (pending, active, completed, cancelled) and produces a transcript.

See: Channels, consent, identity.

Channels

The delivery surface for a run: voice, chat, or embedded. One channel per run.

See: Channels, consent, identity.

Analysis

LLM-generated summary, themes, and sentiment extracted from a completed run. Can be re-triggered.

See: Decision-ready output.