Choose a Zep solution | Zep Documentation

Start with the information that the agent needs. Then identify who owns that context. The three solutions use the same lifecycle with different graph scopes and retrieval paths.

Solution	Use it when	Implementation path
Enterprise context graphs	Agents need shared context about projects, products, operations, or a business domain.	Create a Context Graph with `graph_id`, add business data and documents, then retrieve with `graph.search`.
Agent memory	An agent needs context from one user’s previous conversations, activity, and changes.	Create a user and threads, ingest messages and user business data, then retrieve with `thread.get_user_context`.
Customer and account context	An agent needs personal context and shared customer or account records for the same task.	Keep personal data in a user graph, keep shared data in an account Context Graph, and combine authorized retrieval results.

Combine solutions when a task requires multiple scopes

An agent can retrieve from more than one authorized Context Graph. For example, a support agent can use:

An account Context Graph for contracts, cases, and product history.
A user graph for the current person’s conversations and preferences.
An enterprise Context Graph for support policies and runbooks.

Keep each data set in the graph that matches its subject and access boundary. Combine the retrieved context in the application.

Next steps

Use Architecture patterns to select the graph, ingestion, and retrieval model.
Use Working with Context to implement the ingest, shape, and retrieve lifecycle.
Use Governance to control access and review activity.