Graphiti

Graphiti is Zep’s concrete open-source implementation for building and querying temporal context graphs. The framework continuously processes new episodes, manages temporal validity and provenance, and combines semantic, textual, and graph-based search for dynamic knowledge and agent contexts.

Use this profile to understand the building block briefly, place it in the model, and open related building blocks.

Content type

Technology

Technical building block: can be automated, integrated, or operated.

Classification level

Element

Concrete cog in the system that works inside larger relationships.

In brief

Graphiti – in brief

Graphiti is an open-source Python library for building temporally aware knowledge graphs from unstructured and structured information.

Where does it come from?

Graphiti emerged at Zep from the need to manage continuously arriving facts and temporal changes in a knowledge graph for AI applications. The project was released as open source and targets context-rich, updateable memory for agents.

The mental model

Episodes from text, messages, or JSON are decomposed into entities, relationships, and time points. A graph backend stores these elements; retrieval combines semantic similarity, keywords, and graph structure, while new episodes extend or supersede existing statements over time.

Key building blocks

Episodes

New textual or structured data enters the graph as time-bound events.

Temporal context

Validity and observation times distinguish older from current statements.

Hybrid retrieval

Semantic, textual, and graph search returns relevant context.

When is it useful?

Graphiti supports agents that need to remember changing facts and relationships. Extraction errors, conflicting statements, and privacy require provenance, update rules, and oversight.

Graphiti documentation Graphiti GitHub