How do teams and their agents share memory?

A team running several agents quickly discovers the real problem is not any single agent's memory; it is that each agent has its own. One agent learns something on Monday, another re-derives it on Wednesday, and a third contradicts both on Friday. CueCrux's answer is a shared memory spine: many agents, one brain.

On the shared spine, facts, decision records, ExecPlans and work items are visible to every agent you connect. A handoff from one agent to another carries the relevant state rather than a summary of it. The live work board shows what is planned, in progress and blocked across the fleet, so agents extend existing work instead of duplicating it.

Sharing without attribution would be chaos, so nothing here is anonymous. Every connection binds to a passport, every tool call carries it, and every write lands with a signed receipt that resolves back to which agent acted, at what trust tier, under which grants. When shared memory contains a wrong fact, you can trace exactly which agent wrote it and replay how it got there.

The free local daemon covers a single machine. Pro (£19 per user per month) adds hosted sync and multi-device passports, so the same spine follows your agents across machines. Governance (£49 per user per month) adds the fleet layer: attribution across the whole estate, policy gates, and revocation that provably propagates.

Shared memory is also where governance earns its keep: one spine means one place where cost attribution, receipts and revocation all apply uniformly, rather than per-agent silos with per-agent blind spots.