Doesn't Swarm already pass context between agents?

Within a conversation, yes, through handoffs and messages. But that context evaporates when the run ends and never crosses to another swarm. Glen is durable, org-shared memory that persists across runs and agents.

How do Swarm agents use Glen?

Add Glen as an MCP tool. Any agent, a triage router or a specialist, calls it to read relevant context and record what it learned in a single round trip.

Shared memory for OpenAI Swarm

OpenAI Swarm makes multi-agent orchestration feel light: you define agents, hand off control between them with simple routines, and let conversation context flow as work moves from one agent to the next. That lightness is the point, but it is also the gap. Swarm passes context within a conversation through handoffs and messages, yet nothing persists once the run ends, and one swarm never learns from another. Every triage agent, every specialist, starts each conversation blind to what prior swarms already figured out. Glen, shared memory for AI agents, gives your Swarm agents a durable, shared brain as a single MCP tool, so the knowledge a handoff produces outlives the conversation.

Connect Glen as an MCP tool and any agent in a Swarm can call it to retrieve relevant long-term context and record new facts in one round trip, independent of the conversation-level context that flows through handoffs. A triage agent can pull what the organization already knows about a caller or a problem before routing; the specialist it hands off to can write back the resolution, so the next swarm that sees a similar case starts with that knowledge instead of rediscovering it. You stop relying on the conversation buffer as your only memory and stop accepting that everything a swarm learned is gone the moment it returns.

Because Glen is org-scoped, that memory is shared across every swarm, every routine, and every other agent in your organization, not bound to a single multi-agent run. Swarm is deliberately minimal about state, it carries context forward during a conversation but persists nothing, so durable cross-run memory is exactly the piece it leaves to you. Glen fills it: one swarm records a fact, a different swarm or even a different framework reads it later. As a standard MCP server, Glen is also readable from Claude Code, Cursor, or any other MCP client, so the lightweight agents you build in Swarm share knowledge with the rest of your stack. Wire it in once over OAuth or an API key and the handoffs start compounding into memory.

FAQ

Doesn't Swarm already pass context between agents?: Within a conversation, yes, through handoffs and messages. But that context evaporates when the run ends and never crosses to another swarm. Glen is durable, org-shared memory that persists across runs and agents.
How do Swarm agents use Glen?: Add Glen as an MCP tool. Any agent, a triage router or a specialist, calls it to read relevant context and record what it learned in a single round trip.