Pluribus ships shared agent memory

The interesting move here is not “memory” in the vague RAG sense; it’s turning prior decisions into a governed, shared system of record for agents. That’s a sharper bet than bigger context windows because it targets the failure mode where agents forget constraints, not just facts.

• –MCP plus REST lowers integration friction, so the same memory layer can sit behind editors, custom agents, and service workflows
• –The shared global memory pool makes multi-agent coordination practical instead of leaving each agent in its own silo
• –Pre-change enforcement is the real differentiator: recall alone is useful, but blocking violations of prior decisions is what makes memory operational
• –Postgres-backed storage keeps the system familiar for teams that want durability and operational control without adopting a bespoke memory stack
• –The space is getting crowded, so Pluribus will need to prove its governance model is better than simpler “persistent context” servers