Why not just build a memory layer myself?

You can, but every project then re-implements a store, embeddings, a write path, and retrieval, and the result is siloed. Glen gives you durable, org-shared memory as one MCP integration, reusable across every agent you ship.

Does it work with my framework?

Yes. Glen is a standard MCP server, so any framework or custom loop that speaks MCP, LangGraph, OpenAI Agents SDK, your own, can read and write the same shared memory.

Shared memory for AI engineers

AI engineers ship the systems that put models into production: the agent loops, the retrieval pipelines, the tool integrations, the evals that keep it all honest. Building memory for those systems is a recurring tax, every project ends up reinventing a vector store, an embedding pipeline, a write path, and a retrieval layer just so an agent can remember anything beyond a single run. Worse, that memory is almost always siloed to one application, so an agent you built last quarter shares nothing with the one you are building now. Glen, shared memory for AI agents, gives the agents you build a durable, org-wide memory as a single MCP tool, so you stop rebuilding the memory layer for every project.

Connect Glen over MCP and the agents you build gain one capability that retrieves relevant long-term context and records new facts in a single round trip. Instead of standing up and maintaining a memory subsystem per project, choosing a store, wiring embeddings, designing a write path, handling retrieval, you point your agent at one tool and it reads and writes durable memory. That collapses a substantial chunk of agent infrastructure into a single integration, and it works the same way whether you are building in LangGraph, the OpenAI Agents SDK, a custom loop, or anything else that speaks MCP.

Because Glen is org-scoped, the memory layer is shared across every agent and every application you ship, not bound to one codebase. An agent in one system records a durable fact; an agent in a different system, built by a different teammate, reads it. That is the part a per-project memory store never gives you: knowledge that compounds across your whole portfolio of agents instead of fragmenting into silos. As a standard MCP server, Glen also slots in next to the other servers your agents use, holding the durable knowledge none of them persist, and the same memory is readable from Claude Code, Cursor, or any MCP client, so your production agents and your own dev workflow draw on one store. Wire it in once over OAuth or an API key and let the memory compound while you ship the rest.

FAQ

Why not just build a memory layer myself?: You can, but every project then re-implements a store, embeddings, a write path, and retrieval, and the result is siloed. Glen gives you durable, org-shared memory as one MCP integration, reusable across every agent you ship.
Does it work with my framework?: Yes. Glen is a standard MCP server, so any framework or custom loop that speaks MCP, LangGraph, OpenAI Agents SDK, your own, can read and write the same shared memory.