Platform / agent infra engineers
Swarm & multi-agent
Three agents wrote conflicting plan tiers. Who won — and can you prove it?
Press Play — watch the same writes hit ordinary memory and CLS++ side by side.
Ordinary memory
CLS++ memory
nothing superseded yet
The problem
Your swarm shares one customer record. Triage writes Pro, billing writes Free, retention writes Enterprise. With last-write-wins memory, the last agent to finish silently owns the truth — and the next agent has no idea the other two ever existed.
Last-write-wins shared memory keeps no history of a belief change. If a wrong write lands last, the next agent acts on silently-corrupted state and cannot tell.
How CLS++ handles it
CLS++ turns shared memory into an append-only, lineage-tracked ledger. Every belief change is attributable to the agent that made it, versioned, and replayable. The current belief is always clean; every superseded belief is archived — never deleted — and linked by lineage.
Under the hood — the real script
Three agents write conflicting beliefs about one shared customer record, then a fourth agent reads it. No server, no API key.
python sales/demo/swarm_memory_demo.py[ triage-agent] writes 'pro' — Customer record loaded: plan tier = Pro.
[ billing-agent] writes 'free' — Saw a failed renewal webhook; assumed downgrade.
[retention-agent] writes 'enterprise' — Payments API confirms upgrade to Enterprise.
NAIVE shared memory:
change history: NONE — overwritten, unrecoverable
CLS++ memory:
CURRENT : 'enterprise' v3 by retention-agent
ARCHIVED : 'pro' v1 by triage-agent superseded -> lineage_to_current
ARCHIVED : 'free' v2 by billing-agent superseded -> lineage_to_current
contradictions logged: 3In a swarm, lineage isn't a nice-to-have — it's how you keep shared state from lying to every agent that reads it next.