What is it?

The Data Explorer agent is a read-only AgentFlow assistant that lets you search, filter, and explore your Reltio data using natural language prompts. It focuses on entity details, relationships, match analytics, tenant metadata, and lineage — without modifying any records.

You can ask questions like "How many customers do we have in New York City?" or "Show me the merge history for this entity", and the agent translates them into governed API calls, always respecting your role-based access and masking rules.

For more information, see Data Explorer.

Attribute filtering and counts

✅ Prompt: Find HCP profiles with more than 10 Addresses.

Why it works (with the right configuration): This prompt assumes that Addresses is modeled as a multi-valued attribute or reference collection on the HCP entity. In that case, the agent can translate it into a filter such as equals(type,'configuration/entityTypes/HCP') and gt(attributes.Addresses.count,10). If your tenant stores addresses differently (for example, nested under Location entities), adjust the attribute path and use custom instructions to give the agent concrete filter examples for your model.

⚠️ Prompt: Give me HCPs with more than 10 address records.

Issue: Using address records instead of the exact label Addresses can cause the agent to misinterpret or miss the attribute, especially when the data model exposes a collection named Addresses. If results look empty or incorrect, rephrase the prompt to match the schema label or ask the agent which attribute name it’s using for addresses.

✅ Prompt: How many HCP profiles have more than 10 Addresses?

Why it works (with limits): The agent can attempt to scan entities and estimate counts, but for very large datasets it may hit scanning limits or time out. If that happens, narrow the scope (for example, focus on a region or specialty) or use Reltio UI advanced search for exact counts, and rely on the agent for exploration and pattern discovery rather than bulk statistics.

Data model discovery

✅ Prompt: What’s the data model for my tenant?

Why it works: The agent summarizes entity types, relationship types, graph types, and groupings in your tenant, and explains primary use cases for each. In a single response, you move from “I don’t know this tenant” to a clear understanding of the entities, hierarchies, and interactions the environment supports.

✅ Prompt: Tell me about relationships in my tenant.

Why it works: The agent groups relationships into categories such as core entity relationships, organizational hierarchies, product relationships, territory structures, and interaction types. This makes it easier to understand how entities are connected and which relationship types support specific business workflows.

✅ Prompt: Tell me about your capabilities, including a list of tools.

Why it works: The agent responds with a human-readable overview of what it can do (for example, entity search, relationship exploration, match and merge analytics, audit and metadata queries) and lists the underlying tools it uses. This helps both business users and technical owners understand the governed APIs and functions behind the natural language experience.

Audit and merge history

✅ Prompt: Show me the merge history for entity 00Ei8qS.

Why it works: Providing a specific entity ID lets the agent call audit and match analytics tools for that profile. It can return a timeline of merges and unmerges, the contributing source systems, and associated match scores so you can see how the current golden record was formed.

✅ Prompt: Does entity 00Ei8qS have duplicate addresses?

Why it works: For a single profile, the agent can inspect the Addresses collection, identify potentially duplicated addresses, and summarize where values appear to be repeated or inconsistent. This supports focused data quality checks on high-value entities.

✅ Prompt: Analyze this HCP profile for data quality issues and summarize key findings. The profile is at this URL: entities/00Ei8qS.

Why it works: Asking for a summary and calling out data quality explicitly encourages the agent to highlight missing identifiers, duplicate values, and outliers (for example, unusually high numbers of addresses), along with suggestions for next steps.

Relationship queries (complex)

⚠️ Prompt: List all HCPs affiliated with The Homestead at Soldiers and Sailors Memorial Hospital.

Issue: Long, natural-language names can be ambiguous, and the agent may need to infer both the correct entity and the relationship type. This can result in stalled or incomplete execution, especially if there are multiple hospitals with similar names or multiple affiliation relationship types.

✅ Prompt: Find the entity for “The Homestead at Soldiers and Sailors Memorial Hospital” and show me its entity ID.

Why it works: Breaking the workflow into steps helps. First, you confirm you have the correct hospital entity and its ID before asking the agent to traverse relationships.

✅ Prompt: Show me HCPs affiliated with this hospital using Contact Affiliation relationships (entity ID XYZ).

Why it works: Providing both the hospital’s entity ID and the relationship type gives the agent a precise starting point and traversal path. It can then query the relationship graph and list the affiliated HCPs with fewer ambiguities.

⚠️ Prompt: Which HCPs are affiliated with more than 3 hospitals in New York?

Issue: Combining a complex relationship condition (more than 3 hospital affiliations) with a broad geographic filter (New York) can force the agent to scan large portions of the graph. This may exceed scanning limits or lead to partial results.

✅ Prompt: Find HCPs in New York who are affiliated with at least 3 hospitals, and start with a small sample I can review.

Why it works (with the right configuration): This pattern only works if hospital affiliations are modeled as a reference attribute on the HCP entity and can be counted. The agent needs to run a filter similar to equals(type,'configuration/entityTypes/HCP') and gt(attributes.HCOAffiliation.count,2), where HCOAffiliation is the reference attribute. If your tenant represents affiliations through separate relationship entities or nested structures, update the attribute path in the filter and reflect that in your prompts or custom instructions so the agent can use the correct syntax.

Best practices