Enterprise context graphs
An agent can fail a task when the required business context is split across documents and operational systems. An enterprise Context Graph combines related data into shared context that authorized agents and applications can retrieve.
Use this solution for projects, products, policies, operations, cases, and other business domains that are not owned by one application user.
Choose the graph scope
Create a Context Graph with a stable graph_id for each subject and access boundary. Do not put unrelated subjects or data with different access requirements in one graph.
Examples include:
- One graph for a project or customer account.
- One graph for a product and its support knowledge.
- One graph for an operational domain, such as incident response.
- Separate graphs for business units that have different access policies.
Implementation path
Create and describe the Context Graph
Create a graph with a stable ID, name, and description. The description helps an agent or application select the correct graph.
Ingest business data and documents
Add text, JSON, messages, or document chunks with graph.add. Use zep-ingest for a backfill or a recurring bulk import.
Shape the graph for the domain
Use custom entity and edge types when the application must retrieve domain-specific objects and relationships.
Retrieve context
Use graph.search for a specific graph. You can also combine selected search results with a custom retrieval method.
Apply governance
Use policy-based access control to limit graph access. Use source episodes for provenance and API Logs for supported request-audit workflows.
The Give Your Agent Domain Knowledge guide contains a complete SDK example for graph creation, ingestion, and search.
Application responsibilities
Zep returns context from the graphs that the caller can access. Your application must select the correct graph scope, treat retrieved data according to its trust level, and authorize any external action separately.