Starter · Part of Knowledge Management
Knowledge Flows — visual authoring + agent execution
sub-spec 32A
Decision-tree authoring + agent-facing execution. Some knowledge isn't a document; it's a decision tree ("refund? did they receive it?"). Flows turn a multi-step SOP into a one-question-at-a-time interactive walk with completion tracking.
Knowledge Flows authoring — decision-tree workflows alongside articles in the same knowledge surface. Each flow carries title, summary, state (draft / published / archived), category, and last-updated. Editors filter by state and search; New flow opens the visual authoring canvas. Two flows shown: PCI cardholder verification and refund-eligibility decision.
1 / 2
For the operator
KM lead builds a flow at /knowledge/flows/new — nodes via form, edges wired, terminal coverage validated, graph preview live. Agent on a call opens the flow at /knowledge/flows/[id]/run, answers one question at a time, hits the terminal outcome. The whole session is captured for QA (Quality Assurance: the program that scores and reviews agent interactions.) review. The next agent runs the same canonical flow.
Business impact
Decision logic that lives in supervisor heads or laminated cheat-sheets fails on day 30, day 60, day 90 — and especially when policy changes. Canonical flows + session capture turns the variability in how agents handle the same decision into observable, coachable, improvable data. The 'why did Mike escalate but Jane resolved' question becomes a query, not an argument.