Salesforce · Data 360 · Agentforce

Data Graphs in Data Cloud (Data 360): How They Work, Types, Limits and Use Cases

A data graph is the step that turns a unified customer profile into something an agent, a prompt or a website can read in one call. Here is what data graphs really are, how standard, streaming and real-time graphs differ, what they cost, the limits that shape your design, and what changed in 2026.

Key Takeaways

  • A data graph is a precalculated, read-only view in Data 360 (formerly Data Cloud). It combines one primary data model object and its related objects and insights into a single JSON record, so apps and AI can read a whole customer in one call.
  • It is not a graph database. There are no nodes, edges or GraphQL: you query data graphs through the Data Graph API, Apex, an invocable action, Prompt Builder and Agentforce.
  • Standard graphs refresh on a schedule, streaming graphs update within minutes of a change, and real-time graphs give sub-second access but need the Real-Time SKU.
  • You pay to build a graph and again to query it. Build cost follows the source rows processed, so lean graphs with filters and sensible refresh intervals cost less.
  • Design carefully up front: once a graph is built you can add to it but not remove fields or objects, and you cannot change its refresh type.

What are data graphs in Data Cloud? They are materialized views that Data 360 builds from your harmonized data model objects (DMOs). You pick a primary object, such as Unified Individual, add the related objects, fields and calculated insights you need, and Data 360 flattens them into one JSON record per primary record. Because the record is ready in advance, Agentforce, Prompt Builder, Salesforce Personalization and your own apps can read it almost instantly.

We first published this post in February 2025. This version is rebuilt from Salesforce Help, developer documentation, Trailhead and the 2026 release notes, checked in October 2026. Salesforce renamed Data Cloud to Data 360 on October 14, 2025, and says the functionality is unchanged. Parts of the product, such as the "Data Cloud Architect" permission set, still use the old name. This guide is for Salesforce admins, architects and RevOps and marketing operations leaders at growing and mid-market companies who are planning Data 360 or already run it.

Section 01 · Definition

What Is a Data Graph in Salesforce Data 360?

A data graph is a saved, precalculated combination of Data 360 objects, stored as one JSON record for each primary record. Salesforce describes it as bringing together structured data from your data model objects into easy-to-use views. The data is prepared before anyone asks for it, so a request for one customer returns the profile, contact points, orders and insights in a single response.

The developer guide puts it precisely: a data graph "combines and transforms normalized table data from data model objects (DMOs) into new, materialized views of your data", and because the data is precalculated, queries respond in near real time (Data Graph API overview). Salesforce Help adds that the result is a read-only record holding all the combined data in one JSON blob (Data Graphs in Data 360).

Data Cloud Data Graphs builder for a data graph named Real-Time Personalization Model. The left panel lists seven selected data model objects under the primary object Unified Individual: Individual Identity Link, Individual, Sales Order, Sales Order Product, Primary Product and Einstein Email Engagement Scores, with field counts. The right panel shows the related objects as a map coloured by category: profile, engagement and other.
The data graph builder: Unified Individual is the primary object, and each related object shows how many of its fields are selected. Captured from Salesforce in 2025, when the product was called Data Cloud.

Is a data graph a graph database?

No. The name suggests nodes, edges and graph query languages, and many articles, including the earlier version of this one, described it that way. Salesforce's documentation does not. A data graph is a denormalized view of relational DMOs. You read it by primary key or lookup key through the Data Graph REST API, Apex ConnectApi methods, an invocable action or Prompt Builder, not with GraphQL or graph traversal. The relationship map in the builder above is a design aid, not a separate graph engine.

Data 360 featureWhat it producesUse it when
Data model object (DMO)Normalized, harmonized tables such as Individual or Sales OrderYou map and store source data in a standard shape
Calculated insightMetrics computed across objects, such as lifetime valueYou need an aggregate measure per customer or account
SegmentA list of profiles that meet criteriaYou target or activate a group of people
Data graphOne prejoined JSON record per primary recordAn app, prompt or agent needs a full view of one record, fast
Section 02 · How It Works

How Do Data Graphs Work?

You choose one primary DMO and the related objects and fields you need. Data 360 then flattens them into a nested JSON record for every primary record and refreshes it on the schedule you set. Each data graph lives in one data space and has exactly one primary DMO. That choice decides which related objects you can add.

Step 1 · Unify

Source Data Becomes Harmonized DMOs

Data streams from Salesforce clouds and other systems are mapped to DMOs, and identity resolution creates Unified Individual or Unified Account profiles.

↓ choose a primary DMO ↓
Step 2 · Shape

The Data Graph Defines the View

Pick the primary DMO (any category except Engagement; Profile is most common), then related Profile, Engagement or Other DMOs, calculated insights and streaming insights, with filters and child-record limits.

↓ build and refresh ↓
Step 3 · Serve

One JSON Record Per Customer

Each record is read-only and queryable. Prompt Builder, Agentforce, Salesforce Personalization, Marketing Cloud Next, flows, Apex and external apps read it by key.

Salesforce Help shows the resulting structure with a Unified Individual primary object, its identity links and individuals, and nested arrays for contact points and orders (data graph data structures). A simplified, illustrative record looks like this:

{
  "unifiedIndividualId": "UI-000123",
  "firstName": "Dana",
  "contactPointEmail": [
    { "emailAddress": "dana@example.com" }
  ],
  "salesOrders": [
    { "orderNumber": "SO-88412",
      "items": [ { "productId": "P-17", "quantity": 2 } ] }
  ]
}

The keys are simplified here; in a real graph they are the API names of your DMOs and fields, and the shape follows the objects you selected. Primary keys, key qualifiers and the foreign keys a relationship needs are selected for you automatically. Calculated insights can only be added at the root of the graph, and only when they are based on a DMO that is also in the graph and use its primary key as a dimension (adding calculated insights).

Getting the unified profile right comes first. Our guide to merging Data Cloud with your CRM covers the ingestion, mapping and identity resolution that every data graph depends on.

Section 03 · Types

What Is the Difference Between Standard, Streaming and Real-Time Data Graphs?

They differ in how fresh the data is, what it costs and what they allow. Standard graphs refresh on a schedule, from every 30 minutes to monthly. Streaming graphs are a refresh type of standard graph that updates within minutes of a source change. Real-time graphs serve data with sub-second latency for live sessions, and need the Real-Time SKU.

Standard (scheduled)StreamingReal-time
FreshnessEvery 30 minutes, 1 hour, 4 hours, daily (default), weekly or monthlyWithin minutes of a change, "not in real time"Sub-second for active sessions; the lakehouse copy refreshes hourly
How it updatesFull refresh each run, plus incremental runs where enabledOnly the affected records; one full load at creationContinuously, from real-time events
Relationships1:1, 1:N and N:1No N:1, no filter conditions1:1 and 1:N only for new graphs; joins use the parent's primary key
LicenceData 360Data 360Real-Time SKU
Typical usePrompt grounding, sales and service context, reporting feedsAgents and personalization that need current profilesWebsite and app personalization, real-time segments, insights and data actions

Sources: create a data graph, streaming data graphs, N:1 restriction in real-time graphs and the Data Graphs Trailhead module. The create wizard labels the standard type "near real-time"; Trailhead and the limits page call it "standard". They are the same thing.

Choose real-time when a decision happens inside a live session, such as the next web page, an in-app offer or a service call in progress. Real-time identity resolution, segments, insights, data actions and Salesforce Personalization all require a real-time data graph first (real-time capabilities).

Stay with standard or streaming when minutes or hours of delay are acceptable, which covers most prompt grounding, sales email drafting and case summaries. They avoid the Real-Time SKU and its sub-second usage charges, and streaming removes the need for a daily full rebuild.

Section 04 · Use Cases

Where Are Data Graphs Used in Salesforce?

Wherever something needs a full, current view of one customer or account without running joins at request time. In practice that means grounding prompts and agents, personalizing websites and messages, and serving profiles to flows, Apex and external applications. The same graph can feed several of these at once.

Prompt Builder grounding

Generative AI

Best for: sales emails, case summaries and record overviews that should reflect orders, engagement and insights, not just CRM fields.

Insert an active data graph from the resource picker and it is added to the template as DataGraph:InputName:DataGraphName, then previewed as JSON. Sensitive data is masked before the prompt reaches the model. A few rules matter at design time: only whole graphs can be inserted, the template's input object must map to the graph's root DMO or connect to a Unified Profile DMO at the root, and graphs work with DMOs fed by CRM data streams for standard and custom objects (ground prompts with data graphs). Salesforce's worked example uses a "Contact Overview" graph in a sales introduction email (sales email example).

Prompt Builder template named Summarize a Customer's Order Status in advanced mode. The prompt asks for delivery time, delivery date and tracking number. The Insert resource menu is open on Merge Field Objects with options Sender, Recipient, Organization, Data Cloud and Other, and the cursor is on Data Cloud. Below, the playground shows the grounded prompt and the generated response.
Grounding a Prompt Builder template with Data Cloud data: the Data Cloud option in the resource picker is where data graphs are inserted. Captured in 2025; current menus may differ.

Agentforce

Agents

Best for: service and sales agents that need a customer's orders, cases and status in context.

Agents reach data graphs through actions built on prompt templates, flows or Apex (options to ground an agent). For agentic memory, you can also create an additional real-time data graph in the agent's data space and associate it with the agent, so the GetContext API returns structured data with each call (real-time graph for GetContext). With granular governance, several users and agents can share one graph and each sees only the fields they are authorized to see. For the wider picture, see how Agentforce works.

Salesforce Personalization and Marketing Cloud Next

Marketing

Best for: web, app and message personalization driven by current behaviour.

Salesforce Personalization uses profile and item data graphs, and reads the current profile from the real-time profile data graph when it makes a decision (Personalization developer guide). Recommenders need a profile graph and an item (catalog) graph. In Marketing Cloud Next, content loads a profile data graph by default, available in Handlebars as $dataGraph (Handlebars data graph guide). Our Marketing Cloud Personalization guide covers the marketing side.

APIs, Apex and Flow

Developers

Best for: custom Lightning components, portals, middleware and external apps that need a profile by ID, email or phone.

  • REST: GET /api/v1/dataGraph/metadata, GET /api/v1/dataGraph/{name}/{id} and lookups with ?lookupKeys=[...], with options for live data, projections (selected fields) and filters (lookup endpoint).
  • Apex: ConnectApi.CdpQuery.getDataGraphData, getDataGraphDataWithLookupKeys and getDataGraphMetadata. For a real-time graph, Apex falls back to the standard graph if the real-time one is unavailable (CdpQuery reference).
  • Flow and agents: the standard invocable action "Get Data Graph Data By Lookup" (cdpGetDataGraphByLookup, API 63.0 and later) (action reference). Spring '26 renamed the automation event for event-triggered flows to "Real-Time Data Graph Record Change".

Live queries return the newest data but have longer latency and use more credits, so default to the precalculated graph and reserve live mode for the moments that need it.

Section 05 · Setup

How Do You Create a Data Graph in Data 360?

Open the Data Graphs tab, start from scratch or from a data kit, pick the type, data space and primary DMO, add related objects and fields, preview the JSON, then save and build with a refresh interval. You need the Data Cloud Architect permission set. A build takes anywhere from 15 minutes to several hours, depending on the size of the objects involved.

1. Start the graph.
→ From the App Launcher, open Data Graphs and click New. Choose Start from Scratch, or Use a Data Kit to install a packaged design. Graphs from managed-package data kits can't be edited.
2. Choose the type.
→ Pick near real-time (standard) or real-time. For real-time, you also set consumption limits on the next screen.
3. Set the properties.
→ Enter the name, API name, data space and primary DMO. The API name and data space can never be changed, so name the graph after its purpose, for example Service_Customer_Context.
4. Add fields and related objects.
→ Select fields on the primary DMO, then add related objects with the plus icon or search. Where more than one path exists, choose deliberately: longer paths add levels to the graph.
5. Filter and limit.
→ On the Filters tab, add conditions (AND logic only) and sort-and-limit rules so each customer carries, for example, the 20 most recent orders rather than every order since 2015.
6. Preview, save and build.
→ Preview the JSON, then Save Draft or Save and Build. Choose the refresh interval or Streaming. You can't switch between streaming and scheduled afterwards.

Steps follow Salesforce Help: create a data graph, apply filters and limit child records. Trailhead's hands-on unit walks through a small example.

Managing a graph after it is built

  • Editing is additive. Drafts can add and remove anything. Built graphs can add objects and fields and change the name and description, but can't remove existing fields or objects. The current version stays active while the edit rebuilds (edit a data graph).
  • Cloning keeps the foundations. A clone inherits the data space and primary DMO; everything else can differ. Use it to make a slimmer variant instead of editing a graph that is in use.
  • Deleting is final and removes the graph's schema and JSON. Objects, fields and insights a graph uses can't be deleted while it references them.
  • Watch two statuses. Since October 2026, each graph shows its Status (Draft, Processing, Error, Active) and its Last Run Status (None, Processing, Success, Failure) separately, and Refresh History shows each run's duration and record counts (data graph status).
Section 06 · Limits

What Are the Limits of Data Graphs?

The limits that most often shape a design are 25 objects and 200 fields per graph, five levels of nesting, up to 1,000 child records per object, 30 days of engagement data, and 200 KB per real-time graph record. Each org can hold 25 standard and 25 real-time data graphs. Plan graphs around a purpose, not around everything you could include.

LimitStandardReal-time
Data graphs per org25 (5 drafts)25 (5 drafts)
JSON records per graph200 million100 million
Size of one graph record–200 KB; performance may degrade above it
Objects per graph25, counting the primary DMO, related DMOs and calculated insights
Fields50 per DMO; 200 fields and measures in total (keys excluded); 5 measures per calculated insight
Nesting depth5 levels below the primary DMO
Child records per object100 by default, up to 1,000
Engagement data100 events and 30 days, whichever is lower30 days
Real-time extras–10 real-time data actions and 20 real-time insights per graph

Source: Salesforce Help, Data 360 limits and guidelines, checked October 2026. Limits change between releases, so confirm them in that article before you commit to a design. One trap is worth knowing: if orders or cases are modelled as engagement DMOs, the 30-day window excludes anything older unless you change the recency filter options, as Trailhead's own example points out.

Section 07 · Cost

How Much Do Data Graphs Cost to Run?

Data graphs are consumption-billed in two parts: building them and querying them. Each graph creates a transform in the background, and its build cost depends on the total source records processed on each refresh, not on the number of output records. Query cost is charged when agents, prompts and APIs read the graph. Usage appears in Digital Wallet.

  • Profile objects process all their rows on every full refresh; engagement objects process all rows in the configured time window.
  • One small change can be expensive. Changing a single field, such as a product category, can update thousands of graph profiles.
  • Activations count double. Records processed for a graph used in an activation are counted twice.
  • Live and federated queries cost more, and a graph that includes an external data lake object can consume data federation credits.
  • Real-time graphs drive their own usage types: sub-second real-time events and sub-second real-time profiles, counted as unique active visitors per month.

Salesforce's billing considerations for data graphs lists the levers: pre-filter data before it reaches the DMO, refresh only as often as the use case needs, narrow engagement windows, add filters, and query on the root DMO's primary key, where lookup time stays flat however large the graph grows. Credit rates depend on your contract, so model them with your Salesforce account team. This is where most Data 360 budgets are won or lost, which is why our Data 360 implementation service starts every engagement with a consumption model before anything is configured.

Section 08 · Best Practices

What Are the Common Data Graph Mistakes, and How Do You Avoid Them?

Most problems come from building one large graph for everyone, refreshing it too often, and discovering design limits after it is live. Because built graphs only accept additions and the refresh type is fixed, the cheapest fixes are the ones made on paper before the first build.

One graph for every consumer. A graph sized for marketing, service and agents together grows towards the 200-field and 200 KB limits and costs more on every refresh. Build purpose-specific graphs instead; the 25-graph allowance exists for that.
Fields you can't remove later. Built graphs accept new fields but never lose old ones. Start with the fields a named prompt, agent or page actually reads, and add more when a real request needs them.
Refreshing faster than the use case needs. Intervals shorter than daily raise cost and can cause "data drift", temporary inconsistencies that the automatic 24-hour full refresh corrects (data drift). Use streaming or real-time where freshness truly matters.
N:1 relationships in the wrong graph type. New real-time graphs block many-to-one joins, and streaming graphs don't allow them at all. Map the relationships before choosing the type.
Prompt templates that can't see the graph. If the template's input object is not the graph root and isn't connected to a Unified Profile DMO at the root, the graph won't appear in Prompt Builder. Design the root around the record the prompt runs on.
Assuming the graph hides data. Graphs inherit tags from their source DMOs, and queries now remove or mask fields a user can't access. If you relied on child-object restrictions to hide a whole graph, apply tags or an explicit deny policy on the output object (granular governance).

A design checklist before the first build

  1. 1

    Name the consumer: which prompt, agent, page or API will read this graph, and what it needs to answer.

  2. 2

    Choose the primary DMO from that consumer's record, and confirm the identity resolution behind it.

  3. 3

    List the minimum fields, the child-record limits and the time window for each related object.

  4. 4

    Pick the type and refresh from the freshness the use case needs, then estimate build and query volumes.

  5. 5

    Build in a sandbox, check the JSON preview and query by primary key, then deploy with a data kit.

Section 09 · Release Notes

What Changed for Data Graphs in 2026?

A lot: 2026 brought streaming graphs, a higher graph limit, projections, refresh history, field-level governance and incremental edits. Several older articles still describe data graphs as daily-only with ten graphs per org. These are the changes from the Summer '26 and Winter '27 release notes that matter most for design.

ReleaseChangeWhy it matters
Spring '26Flow event renamed "Real-Time Data Graph Record Change"Update team documentation and naming conventions
Summer '26Up to 25 data graphs, up from 10Purpose-specific graphs per agent or channel become practical
Summer '26Streaming data graphsMinutes-fresh profiles without a daily full rebuild, billed on updates processed
Summer '26Contact lookup by email or phone on real-time graphsFind a profile without knowing its unified ID
Summer '26Projections and refresh historySmaller API payloads and LLM context; visible run durations and record counts
Summer '26Data graph attributes in streaming DMO activations; default profile graph per data space for PersonalizationSimpler activation filters and Personalization setup
Winter '27Granular governanceOne graph can serve users and agents with different access
Winter '27Separate Status and Last Run Status; reliable redeploy of data kits with graphsClearer monitoring and safer deployments
Winter '27Incremental edits to streaming graphsAdd fields without a full rebuild
Section 10 · Bottom Line

The Bottom Line

Data graphs are where Data 360 stops being a data project and starts serving people and agents. Designed around a named consumer, they give Agentforce, Prompt Builder and Personalization a complete customer in one call. Designed as a catch-all, they become slow, expensive and hard to change. Twopir Consulting designs and builds Data 360 implementations, including the data graphs and Agentforce grounding that sit on top. For the platform basics, start with our Data Cloud growth guide.

  1. 1

    Treat a data graph as a precalculated JSON view, not a graph database.

  2. 2

    Build one graph per consumer, with the minimum fields and child records it needs.

  3. 3

    Match standard, streaming or real-time to the freshness the use case truly needs.

  4. 4

    Model build and query costs before you build, and query by primary key.

  5. 5

    Decide fields, relationships and refresh type up front, because built graphs only grow.

Ready to Start

Design Data Graphs Your Agents Can Rely On

Tell us which agents, prompts or pages need customer context. We will map the primary objects, graph types, refresh intervals and consumption your Data 360 org needs.

Schedule a Discovery Call

FAQ

Frequently Asked Questions

A data graph is a precalculated, read-only view in Data 360, the new name for Data Cloud. It combines one primary data model object, such as Unified Individual, with related objects, fields and calculated insights, and stores the result as one JSON record per primary record, so apps, prompts and agents can read a full profile almost instantly.

No. Despite the name, Salesforce documents data graphs as materialized, denormalized views of relational data model objects, not as a graph database with nodes and edges. You read them by primary key or lookup key through the Data Graph API, Apex ConnectApi methods, an invocable action or Prompt Builder, not with GraphQL.

Standard data graphs refresh on a schedule, from every 30 minutes to monthly, or by streaming within minutes of a change. Real-time data graphs serve sub-second data for active sessions and power real-time segments, insights, data actions and Salesforce Personalization. Real-time graphs require the Real-Time SKU and allow only one-to-one and one-to-many relationships for new graphs.

Standard graphs refresh daily by default, and you can choose every 30 minutes, 1 hour, 4 hours, weekly, monthly or streaming. Every graph also gets an automatic 24-hour full refresh that corrects any data drift from frequent refreshes. A build or full refresh takes from about 15 minutes to several hours, and data stays readable during a refresh.

In Prompt Builder, insert an active data graph from the resource picker; it is added as a DataGraph merge field and sent as JSON, with sensitive data masked. The template's input object must map to the graph's root or connect to a Unified Profile at the root. Agents then use that prompt template, or a flow or Apex action, to read the graph.

Salesforce Help lists 25 standard and 25 real-time data graphs per org. Each graph can include 25 objects, 50 fields per object and 200 fields and measures in total, nested up to five levels below the primary object, with up to 1,000 child records per object. A real-time graph record should stay under 200 KB.

Partly. You can rename it, change the description and add objects and fields, and the current version stays active while it rebuilds. You cannot remove fields or objects from a built graph, change its API name or data space, or switch between streaming and scheduled refresh. Drafts can be changed freely, so finalize the design before you build.

Yes. Building a graph is billed on the source records processed during each refresh, and querying it is billed separately. Records count double when a graph feeds an activation, live and federated queries cost more, and real-time graphs add sub-second usage types. Fewer fields, filters, sensible refresh intervals and primary-key lookups keep costs down.

TC

Twopir Consulting

Salesforce Partner · HubSpot Partner

Twopir Consulting helps growing and mid-market companies solve complex CRM, integration, and business system challenges. We implement, rescue, optimize, and integrate complex CRM and revenue systems, including Data 360, Agentforce and Marketing Cloud, for 500+ clients over 12+ years.