Lola is the agent I actually live with. She is a chief of staff, not a chatbot: she triages email, keeps semantic memory of what matters, manages tasks, and handles content security. She runs on her own host, local-first, and she is Lola because she has her own memory and context — that is the rule I hold for her. An animated face is animation only; the words come from her.

The reason Lola works is the same reason most assistants do not: she persists. A session that forgets everything is a demo. A chief of staff that remembers what you decided, who you owe an answer, and what you said last month is infrastructure. That is what the memory layer is for, and it is why her host matters — her context lives where I control it.

The Lolabots Factory generalizes the pattern. At lolabots.com you pick a role from a catalog of over 300 specialist templates, customize it, and deploy it — roughly a minute from choice to a working agent, free, no signup. The Factory builds agents; it does not connect agents that already exist. That separation matters: building is one layer, messaging is another, and each should be good at its own job.

If you want to try the pattern, Lola herself is the fastest start: clone the lolabot repo, run the setup script, and you have a working chief of staff — email, memory, tasks, and security included — in seconds. Then change her. The point of a template is not to keep it; it is to have something real to edit.

Links: lolabots.com, github.com/23blocks-OS/lolabot