How to pull Zep Memory + Knowledge Graph data during an active ConvAl (Widget) session?

[cedar kettle]

I’m trying to integrate Zep Memory and the Zep Knowledge Graph into my ElevenLabs Conversational AI Agent (Widget) so the agent can access:

User history & session context from Zep Memory

Facts & structured knowledge from the Knowledge Graph

…while the conversation is still ongoing, not just before or after.

Why I need this:
I want the agent to dynamically adapt responses based on both user-specific history and relevant knowledge from Zep mid-conversation.

What I’ve tried:

MCP integration → ConvAI doesn’t accept it.

Webhook approach → only works at the start or end of the call, not for multiple retrievals during the conversation.

My question:
What’s the best approach right now to achieve live, mid-conversation lookups from an external memory + knowledge graph?
for context: https://help.getzep.com/overview

Has anyone here done this, or does ElevenLabs have an official recommendation?

How to pull Zep Memory + Knowledge Graph data during an active ConvAl (Widget) session?

[warm knoll]

Were you able to devise a way to store and retrieve Zep data inside Eleven Labs agents?

[mild tangle]

For live mid-conversation lookups, you’ll need a real-time layer between ElevenLabs and Zep that queries Memory and the Knowledge Graph per user input and injects relevant context back into the conversation dynamically. Using session-aware embeddings or a lightweight API proxy can let the agent adapt responses on-the-fly without restarting the session. Have you considered building such a proxy, or are you looking for an official ElevenLabs-supported approach?
@cedar kettle