What is Shared agent memory?

Shared agent memory is the team-scoped design of agent memory, where the store is not private to a single user or session but common to a group, typically an organization. In the usual per-user model, each engineer's agent keeps its own memory, so a fact one person's agent discovers, a deployment quirk, a naming convention, why a past approach was abandoned, stays trapped with that one agent and everyone else's agent rediscovers it the hard way. Shared memory removes that wall: every member's agent reads and writes the same store, so an observation recorded by one is immediately retrievable by all, and the team's collective context compounds instead of fragmenting. The design challenge is scoping, who can see and contribute to which store, and provenance, tracking which member or agent recorded a given fact, while keeping access simple. Glen is built around this idea: memory stores belong to the organization, every member can read and write every store the org owns, and the whole thing is reached through one MCP tool, so a fact learned by any teammate's agent becomes shared knowledge across the team.