A single connected object rarely holds just one kind of thing — a folder of files, a table of tickets, or a feed of messages usually mixes several record types together. Discover & Enrichment is the wizard step that lets you use AI to sort those records into meaningful groups and enrich them, before they ever reach the mapping step.
Overview
Discover & Enrichment sits between choosing which objects to read and mapping those objects onto graph nodes. It goes beyond simple field-to-field mapping: it lets AI look inside unstructured or semi-structured content — free text, attachments, loosely typed records — identify what each record actually is, and add information to it. "What you build here feeds the Map step next," so your choices shape what you see when you map. This page covers only discovery and enrichment; see the source connection setup guide for connecting a source.
Turning it on: the toggle
Each object you selected gets its own Discover & Enrich specific entities toggle. Turning it on sorts records into entities and enriches each group differently, rather than treating every record the same way. Where it starts depends on the connector — some apps support discovery by default — but you can flip it either way:
Off: a single, flat Enrich All Records list. Every record runs through the same enrichment steps.
On: a Discover entities with picker followed by an Entities list. Discovery runs on every record to classify it, and the entities list splits records into groups by condition, enriching each differently.
Flip the toggle before configuring anything — switching it discards whatever you've built for that object, immediately and without a confirmation prompt. The step is optional per object, but once started, it must be finished before you can continue.
Agent-based and Automation-based methods
Whether you're configuring the discoverer or an entity's enrichment, you choose one of two methods, then the specific agent or automation to run:
Agent-based. An AI agent receives the full record — no input mapping needed — and classifies or enriches it using its own reasoning. View Schema shows what fields it returns.
Automation-based. A workflow does the extraction instead. Since it expects specific inputs rather than a full record, you map each input to a source property yourself with Map inputs. UnifyApps flags unmapped required inputs and rejects mismatched field types at save time.
Only published agents and automations appear in the picker, and your source always runs the latest published version. If the one you want isn't listed, a shortcut opens the builder so you can publish it and return to your draft. Each entity also needs a name and a Classify When condition built on the discoverer's nprt fields; each enrichment list accepts a single enrichment step.
How discovered entities feed into mapping
Your choices here change what you see in Map next. With the toggle off, you map the object itself, same as without this step. With the toggle on, Map replaces the object with the entities you defined — you map each discovered entity to graph nodes and edges individually. Any fields your enrichment added appear in Map as extra source fields, grouped under the enrichment's name, ready to map onto destination properties alongside the record's original fields. In short: discovery decides how records split, enrichment adds to what each record knows, and mapping wires both into your graph's structure.
Notes
Entity names accept only letters and numbers; the name is also the entity's identity, so renaming one effectively re-creates it.
If a condition references a field your discoverer no longer returns, it's flagged for you to fix or remove rather than silently dropped.
Changing the toggle, or the object's underlying connection, resets this step's configuration — settle on your approach before building out entity conditions.