Shared memory for Vapi

Vapi powers voice AI agents, the assistants that answer calls, qualify leads, book appointments, and handle support over the phone. Each call is a fresh conversation, and while you can pass variables in and log transcripts out, the agent does not actually remember the caller across calls or share what it learned with the rest of your stack. The result is repeated questions and context that lives only in transcripts no one reads. Glen, shared memory for AI agents, gives your Vapi assistants durable, org-scoped memory as a single MCP tool, so a voice agent can recall what it already knows about a caller and record what this call surfaced.

Wire Glen into your Vapi assistant as an MCP tool and the agent can call one tool mid-call that both retrieves relevant long-term context, who this caller is, what they wanted last time, the commitment you made, and records new facts before the call ends. The next call, whether to the same assistant or a different one, starts with that context instead of from scratch. Callers stop repeating themselves, and the knowledge from a call stops dying in a transcript.

Because Glen is org-scoped, the memory is shared across every Vapi assistant and across your non-voice agents too, rather than locked to one assistant or one call. A sales voice agent records a durable fact about a prospect; your support voice agent, or a human-driven agent in Claude Code, reads it next time. That is the part per-call variables and transcripts do not give you: persistence that outlives the call and is shared across channels. And because Glen is a standard MCP server, the same memory your voice agents write is readable from any other MCP client, so phone, chat, and editor-driven work all draw on one source of truth. You connect it once over OAuth or an API key and let each call make the next one smarter.

FAQ

Can a Vapi voice agent really call Glen during a live call?
Yes. Glen is exposed as an MCP tool, and Vapi assistants can invoke tools mid-call, so the agent can retrieve caller context and record observations in a single round trip without ending the call.
How is this different from passing variables into a Vapi call?
Variables are set per call and gone afterward. Glen is durable, org-shared memory that persists across calls and assistants and is readable by your other MCP clients.