Salesforce Data 360 (formerly Data Cloud) grounds Agentforce agents in two ways. Structured context comes from unified profiles and data model objects. Unstructured content, such as cases, knowledge, files and web pages, reaches agents through retrieval-augmented generation (RAG): a search index stores the content, a retriever finds the relevant parts, and a prompt template passes them to the agent.
This guide explains each path, how the Einstein Trust Layer applies, and what to prepare before you ground your first agent. For delivery, see our Data 360 implementation and Agentforce implementation pages.
What does grounding mean for an Agentforce agent?
Grounding is how an agent's answer is tied to your data rather than to what a language model already knows. Trailhead's Agentforce Data Library unit defines it as adding "domain-specific knowledge or customer information to a prompt", with grounding sources that "can include knowledge articles, uploaded files, websites, conversation transcripts, and more."
Salesforce positions Data 360 as the place this context comes from. Its Data 360 page calls it "the essential data foundation for Agentforce" and says its Agent Context Engine anchors Agentforce "in verified business logic and real-time context."
How does Data 360 data reach an agent?
There are three practical paths. Most agents use more than one: a profile to know who the customer is, and RAG to find the right answer.
| Path | What it carries | Data 360 parts involved |
|---|---|---|
| Structured grounding | Customer and business records, such as a unified profile, orders or calculated insights | Data lake objects and data model objects; unified profiles built by identity resolution |
| RAG on Data 360 | Long-form text, such as case descriptions, knowledge and documents | Search index (vector or hybrid), retriever, prompt template, agent action |
| Agentforce Data Library | Unstructured sources, such as knowledge articles, uploaded files and websites | Automated data stream, mapping, search index and its own retriever |
Trailhead's Implement Data 360 for Agentforce unit works through a full example: a search index built from the Case DMO and another from the Unified Individual DMO, retrievers created automatically for each, and an agent action named Answer Questions with Case that uses the prompt template.
How do search indexes and retrievers work?
A search index makes text searchable by meaning. Trailhead's search index types unit says Data 360 "creates meaningful content chunks from the data," then "creates vector embeddings to build a search index." It describes the index types this way:
- Vector search, also called semantic search, matches meaning and "works well for long-form search queries where the users are looking for general information."
- Hybrid search combines vector search with keyword search, which suits questions that include specific terms such as product names, error codes or policy numbers. Trailhead recommends a hybrid index for "the most accurate and relevant results."
- Enriched indexes add metadata and question chunks for higher retrieval accuracy, which Trailhead notes carries added LLM cost.
The retriever connects the index to the agent. Data 360 creates a retriever for each index, which Trailhead calls the bridge "between search indexes and prompt templates." When the agent action runs, "the retriever populates the prompt with the most relevant information." The prompt template then frames how the model uses it.
Why does the unified profile matter for agents?
RAG finds the right content; the unified profile tells the agent who it is talking to. Trailhead notes that "RAG solutions built on Data 360 have context from unified profiles," and its example uses a customer's profile details to tailor a product answer.
That makes identity resolution part of agent quality. If two customers are wrongly merged, an agent may answer one with the other's orders; if one customer is split across profiles, the agent sees half the history. Our guide to Data 360 identity resolution explains how match and reconciliation rules shape the profile.
How does the Einstein Trust Layer apply to Data 360 grounding?
Agentforce prompts pass through the Einstein Trust Layer. Trailhead's Trust Layer module lists secure data retrieval, dynamic grounding, data masking and zero data retention, and says "an audit trail tracks a prompt through each step of its journey."
One detail changes design: the same module says "data masking for LLMs is currently disabled for agents." So the controls that matter most are upstream. Index only the content an agent should use, keep sensitive fields out of indexes and actions, and use Data 360 governance. Trailhead's govern and secure unit says users see only the data in the data spaces they can access, and that policies combine tags, classifications and user attributes at field, object or record level.
What should you prepare before grounding agents in Data 360?
Trailhead's first planning step is to "identify what data is needed for your use case, and define the type of data." In practice that becomes a short checklist:
- Pick one agent job. For example, answering order questions or resolving common cases. The job defines what the agent must know.
- Split structured and unstructured needs. Records and profile fields go through the data model; long-form text goes through a search index.
- Clean the content you will index. Retire outdated knowledge articles and duplicate documents. A retriever will surface whatever the index holds, current or not.
- Index the right fields. Index long-form text fields; keep IDs and short codes as structured filters instead.
- Check identity resolution. Test unified profiles on known customers before an agent relies on them.
- Set access. Decide the data space, permissions and fields each agent may use, since LLM data masking does not apply to agents.
- Build a test set. Collect real questions, including hard ones, and check answers and sources before launch.
To prove this on one agent first, our Data 360 Quickstart can take a search index and retriever to production for a single prompt template.
Frequently asked questions
Does Agentforce need Data 360?
Salesforce calls Data 360 the essential data foundation for Agentforce, and capabilities such as RAG with search indexes and retrievers, and the Agentforce Data Library, run on it. Whether a first agent needs more than CRM records and knowledge depends on the use case.
What is a retriever in Data 360?
A retriever is the bridge between a search index and a prompt template. Data 360 creates one automatically for each search index, and when an agent action runs, the retriever fills the prompt with the most relevant content from the index.
What is the difference between vector and hybrid search indexes?
Vector search matches meaning and suits general long-form questions. Hybrid search combines vector search with keyword search, which helps when questions include specific terms such as product names or codes. Trailhead recommends hybrid for the most accurate and relevant results.
What is the Agentforce Data Library?
It links agents and LLMs to unstructured data such as knowledge articles, uploaded files and websites. Trailhead says it automates pushing data streams to Data 360, mapping data objects and creating a search index and retriever, and each library has its own retriever.
Is data masked when Agentforce uses Data 360 content?
Not by the LLM data masking feature. Trailhead's Einstein Trust Layer module says data masking for LLMs is currently disabled for agents. Control what an agent can read through the content you index, data spaces, permissions and the actions you give it.
Talk to a Data 360 and Agentforce architect
Techila Global Services is a Salesforce Summit partner, headquartered in Frisco, Texas. Tell us what your first agent should do, and we will map the Data 360 context it needs.