Skip to main content
Everyone/elie-voice

How Voice IntELIEgence works

A voice application is one saved configuration for one use-case. Every conversation, transcript, output and export job belongs to exactly one application, which takes conversations only once it is published.

What an application decides

Each row is a section of Settings inside the application:

What it decidesWhere you set it
Where conversations come fromSources & ImportManual upload, IntelieKonnect import
Which outputs get generatedOutputs & GradingSummary, Action items, Topics & sentiment, Talk-time stats, Grade
What grading scores againstOutputs & GradingGrading rubric, one criterion per row
Whether a human must approveGeneral settingsRequire human review
How long audio and transcripts are keptGeneral settingsRetention · raw audio (days) and Retention · raw transcript (days)
Whether transcripts also feed a knowledge baseKB & Chat

How a conversation flows

A conversation starts when someone selects Upload conversation — it sits in the application's header bar, so it is reachable from every section — or when an import connector brings one in. It is transcribed, then put through the stages your configuration enables: speaker identification only while that stage is on, then PII redaction, classification, analysis, the knowledge base and export. SettingsApplication details draws this as a Pipeline, with any stage you turned off shown as Skipped.

Where reviewing fits

With Require human review on, a processed conversation waits for a reviewer to approve it before it is marked Completed; with it off, finishing processing is enough. Reviewing happens on the conversation itself — Transcript to read it and correct who spoke, Outputs to check what was generated. Only Summary and Action items can be edited; a grade, topics and sentiment are read-only. Approving locks the outputs for good.

Outputs decide which menu items exist

Two items in the left rail are conditional. Topics & Sentiment appears only while the Topics & sentiment output is on, and Grading only while Grade is on — and Grade is off until you turn it on. Turning one off removes the rail item outright rather than greying it out, along with its card on a conversation's Outputs tab.

Every save pins a new version

Saving any Settings section creates a new version of the application — hence the line Changes apply to new conversations only (a new version is pinned). on every section. A conversation keeps the version it started under, so an edit never rewrites conversations already processed; only conversations started after the save use it.

Next

Steps verified against elie-tenant-ui @ origin/dev 2026-08-04 on . Something wrong with this page?