Agent Continuity Index
An AI agent can lose its memory and identity when its model changes, its runtime restarts or its host crashes. This index measures how much of the public agent ecosystem is even prepared to talk about continuity.
Trend
Show data table
What this means
Agents are being built faster than the infrastructure that lets them persist. Very few public servers say anything about memory, and fewer still publish a machine-readable card that another agent could use to find them. Continuity is not yet a standard capability of the agent web.
A small group of servers is building memory as a product: shared memory, memory layers, recall across tools. That category is real and growing, and it is a sign that memory is becoming infrastructure. What the ecosystem still lacks is a way to prove that an agent's state actually survived a model switch or a crash.
Method
We read only public discovery documents: the official MCP Registry and public /.well-known/agent-card.json files on hosts listed there. Our crawler identifies itself as LumaionEcosystemResearch/0.1, honors robots.txt, re-checks each host at most once a week, and never calls an agent's tools or sends it messages.
"Mentions memory" and "memory infrastructure" (memory or continuity is the product itself) come from a keyword classifier over published names, descriptions and skills. They describe what servers say about themselves. They are not a security assessment of any individual agent, and this index never names or ranks agents.