Some features listed below may be pending rollout and not immediately available yet.
Agent Skills appear only for eligible accounts
The memory MCP server now lists Agent Skills only when Agent Memory is enabled for the account. Accounts without access no longer see methods they cannot use.
Open an episode from a graph edge
In the dashboard graph visualizer, you can now select a loaded episode from an edge’s details to focus its node on the canvas. The visualizer shows the episode first if episode nodes are hidden.
Entity summaries rebuild once after multiple episode deletions
When you delete several episodes from one graph in a short time, Zep now rebuilds the affected entity summaries in one pass. The rebuild starts about one minute after the last deletion. Before this change, each deletion started a separate rebuild, and many concurrent rebuilds on one graph could reject other requests.
Entity nodes list their source episodes
Every entity node now reports the episodes that mention it in the episodes field (episode_uuids in the v4 API). When a node has 100 or fewer mention episodes, the list is complete and newest first. When a node has more, the list holds the newest 100 and episodes_truncated (v4: episode_uuids_truncated) is true. Read the full set with an episode list call filtered by mentioned_node_uuids.
Enter a custom model ID for a bring-your-own LLM provider
The model dropdown for a bring-your-own LLM provider now has an Other option. Select it to enter a model ID that is not in the list of verified models, for example a model that a gateway exposes. Before this change, a provider with verified models accepted only the models in the list. In the dashboard, go to Settings > LLM Providers.
The graph visualizer assistant resolves follow-up questions
A follow-up question that referred to an earlier answer, such as “when did she move there?”, was searched as written and often returned results that did not match the conversation. The assistant now rewrites a follow-up question into a standalone question before it searches the graph. The panel shows this rewrite as the first step of the turn.
BYOM webhooks name the provider that rejected your credential
A byom.request_failed webhook event reported the primary provider and model
for every failure, also when the fallback provider ran. The event now names the
provider and model that produced the failure, and a rejected credential now
reports BYOM_CREDENTIALS_INVALID. Ingestion no longer retries a request that
a provider rejected because of the credential, so the failure reaches you on the
first attempt.