> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prophecy.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# What is a knowledge graph?

> Prophecy creates a metadata index to power AI features

Prophecy uses **Knowledge Graphs** to help AI features such as Prophecy Agent and Copilot understand your [data environment](/data-analysis/environment/fabrics/prophecy-fabrics).

A Knowledge Graph is a metadata index that represents relationships between schemas, tables, columns, data types, and other objects in your fabric. The Knowledge Graph contains metadata about your data environment — not the actual data stored in your warehouse. A Knowledge Graph is a metadata index that represents relationships between schemas, tables, columns, data types, and other objects in your fabric. For general background on knowledge graphs, see the [Wikipedia article on knowledge graphs](https://en.wikipedia.org/wiki/Knowledge_graph).

The Knowledge Graph only indexes metadata that the configured warehouse identity is authorized to access.

When you interact with AI in Prophecy, the platform uses the Knowledge Graph to enrich your prompts with metadata context about accessible datasets, schemas, tables, and columns. Prophecy indexes this metadata using both structured search and embedding-based retrieval techniques to help the model identify relevant schema context from natural language prompts and generate more accurate responses and SQL.

## Knowledge graph generation

Each fabric maintains its own Knowledge Graph index of accessible metadata.

The indexer retrieves metadata from all data connections attached to the fabric, using either your identity or a separately configured identity. During indexing, Prophecy retrieves metadata only for the warehouse objects that the configured identity is authorized to access.

You can schedule automatic refreshes or trigger manual indexing to keep the Knowledge Graph current.

<Note>
  Prophecy uses the Knowledge Graph associated with the fabric attached to your project. If you attach to a
  different fabric, AI features will use that fabric's Knowledge Graph.
</Note>

## Agent behavior when data access is disabled

The `AI_DATA_ACCESS_CLUSTER_ENABLED` flag controls whether the V4 agent can access live data. When this flag is set to `false`, the agent operates in a metadata-only mode and is limited to Knowledge Graph operations.

In this mode, the agent can still use Knowledge Graph metadata to search datasets, retrieve schema information, and provide metadata-aware assistance. However, the AI agent cannot perform live data operations such as executing SQL queries, running pipelines, or generating summaries that require warehouse access.

| Capability                        | Allowed | Agent behavior                                                                                          |
| --------------------------------- | ------- | ------------------------------------------------------------------------------------------------------- |
| KG search                         | ✓       | Returns KG-indexed metadata results                                                                     |
| Dataset discovery                 | ✓       | Lists datasets available in the KG                                                                      |
| Schema lookup (via KG)            | ✓       | Returns schema information stored in the KG without querying the source                                 |
| SQL queries                       | ✗       | Explains that live data access must be enabled                                                          |
| Pipeline execution                | ✗       | Explains that live data access must be enabled                                                          |
| Summarization requiring live data | ✗       | Provides a partial response using available metadata and explains that live data access must be enabled |

<Note>
  When data access is disabled, the agent can still provide metadata-aware assistance using the Knowledge Graph, but live data operations require data access to be enabled.
</Note>

## What's next

Configure the [knowledge graph indexer](/data-analysis/ai/knowledge-graph/indexer) to schedule automatic indexing or set up separate authentication credentials.