API Reference Zep v#1.0.0

Copy Markdown View Source

Modules

Zep

A complete Elixir client for the Zep AI memory API.

The Batch API - the recommended way to load large historical datasets (backfills, document collections, archived conversations, migrations) into Context Graphs. Replaces the deprecated Zep.Thread.add_messages_batch/4 and (for large jobs) Zep.Graph.add_batch/3.

A configured handle for talking to the Zep API.

Resolves client configuration from explicit options, application environment, and OS environment variables, in that order of precedence.

Raised when a Zep.Client cannot be built due to missing configuration.

Custom context-rendering templates, referenced by template_id from Zep.Thread.get_user_context/3 and Zep.Thread.add_messages/4.

A structured representation of a failed Zep API call.

Custom entity/edge extraction instructions for a user's or graph's graph - free-text guidance steering what Zep extracts during ingestion (e.g. "always extract monetary amounts as Money-labeled nodes").

Operations on individual graph edges (facts/relationships between nodes).

Operations on individual graph episodes (units of ingested data).

Operations on individual graph entity nodes.

Observations - standalone, timestamped notes attached directly to a user's or graph's graph outside of the thread/episode ingestion flow.

Access to the thread summaries that have been folded into a user's or graph's knowledge graph (as distinct from Zep.Thread.get_summary/2, which returns a single thread's own rolling summary directly).

Low-level request execution shared by every resource module.

Helper for turning a paginated list/2-style function into a lazy Stream that walks every page automatically.

Project-level settings for the API key currently in use (the project the key was issued under).

A single item within a batch, as returned by Zep.Batch.list_items/2.

A batch ingestion job, as created/listed via Zep.Batch.

Item-count progress for a batch job, as returned nested in Zep.Schemas.BatchJob.

A custom context-rendering template, referenced by template_id.

A fact/relationship edge between two nodes in a graph.

An episode - a single unit of data ingested into a graph via graph.add.

A named graph - either a user graph or a standalone/group graph.

Paginated envelope returned by Zep.Graph.list_all/2.

A single chat message, as stored on a thread.

Paginated envelope returned by Zep.Thread.get_messages/3.

An entity node in a graph.

Project-level settings for the API key in use.

Results of Zep.Graph.search/3 - facts (edges) and/or nodes, depending on scope.

An async task/job handle, as returned by long-running operations.

A Zep thread - a sequence of messages between a user and an assistant.

The assembled context block returned by Zep.Thread.get_user_context/3.

Paginated envelope returned by Zep.Thread.list/2.

The rolling, incrementally-generated summary of a thread.

A Zep user - the identity that threads and graph memory are attached to.

Paginated envelope returned by Zep.User.list_ordered/2.

Status polling for asynchronous, long-running Zep operations (e.g. bulk ingestion, pattern detection jobs).

Operations on individual thread messages, addressed by message uuid (as opposed to Zep.Thread.add_messages/4, which appends to a thread).

Raised/returned when a request never reached the Zep API (timeouts, connection refused, DNS failures, etc), as distinct from Zep.Error which represents an API response with a non-2xx status.

Verifies the authenticity of incoming Zep webhook deliveries.