Unified context | Zep Documentation

Use Choose a Zep solution to select the graph scope and implementation path for your task.

Context is the information that an agent needs to complete a task. Unified context combines related information from business systems, documents, and conversations so agents and applications can use it together.

Zep is the unified context layer for enterprise data. Teams use Zep for three main solutions:

Enterprise context graphs combine knowledge about projects, products, operations, and business domains.
Agent memory provides context from previous conversations, activity, and changes.
Customer and account context combines records, conversations, and events for a customer or account.

A Context Graph represents facts, their relationships, and changes over time. The Context Lake manages and serves Context Graphs as one system.

Governance controls who can manage Zep and what context each agent or application can retrieve. Source episodes provide provenance for facts and other graph data. Audit Logs record dashboard activity, and API Logs record supported API request activity.

Retrieval returns selected context from a Context Graph. The application decides how to place that context in a model request and whether an agent can use a tool or take an action.

Flex Plus and Enterprise accounts can also use Observations. These records describe patterns that Zep identifies in graph data.

Concept	Description	Docs
Unified context	Related information from different sources that agents and applications can use together.	Docs
Context Lake	Zep’s system for managing and serving Context Graphs with shared governance controls.	Docs
Context Graph	A temporal graph that represents facts, relationships, source episodes, and changes over time for a user, customer, account, project, or other subject.	Docs
Agent memory	Context from previous interactions, activity, and changes that an agent can retrieve for later tasks.	Docs
Zep’s Context Block	An assembled string that contains relevant context types from a user’s Context Graph.	Docs
Context Types	The distinct primitives Zep produces from a user graph: facts, entities, episodes, thread summaries, observations, and user summary. Choose the type that fits the agent’s task.	Docs
Fact Invalidation	When new data invalidates a prior fact, the time the fact became invalid is stored on that fact’s edge in the Context Graph.	Docs
JSON, text, and message	Data types that you can ingest into a Context Graph.	Docs
Custom Entity/Edge Types	Feature allowing use of Pydantic-like classes to customize creation/retrieval of entities and relations in the Context Graph.	Docs
Graph	A Context Graph addressed with `graph_id`. Use it for shared or domain context that is not owned by one Zep user.	Docs
User graph	A Context Graph that stores context for one application user.	Docs
User	A Zep representation of an application user. Each user has a user graph and thread history.	Docs
Threads	Conversation threads of a user. By default, all messages added to any thread of that user are ingested into that user’s graph.	Docs
Documents	Grouping for episodes of the same source on a graph. Pass `document_id` when ingesting chunks of a PDF, export, or channel.	Docs
Provenance	Source information that connects facts and graph artifacts to the episodes from which Zep derived them.	Docs
Governance	Controls for team access, agent and application access to context, source traceability, and audit activity.	Docs
`graph.add` and `thread.add_messages`	Methods that add business data and chat messages, respectively.	Business data Messages
`graph.search` & `thread.get_user_context`	Low level and high level methods respectively for retrieving from the Context Graph.	Docs Docs
User summary instructions	Up to five instructions that specify how Zep generates a user summary.	Docs
Debug mode	Time-limited project mode for capturing per-episode workflow logs. Flex, Flex Plus, and Enterprise accounts also capture LLM reasoning traces for selected ingestion steps.	Docs
Agent tool	A tool that lets an agent retrieve information from a Context Graph.	Docs