Six rules we build by, and every bot we make is held to them. They're not marketing. They're the spec.
# six rules we build by.
# enforced by architecture, not policy.
No model, anywhere. There is no language model in the app, on the server, or "just for one feature". Your bot's brain is rules, patterns and facts — the same input gives the same answer on the same day next year. When it speaks, it speaks lines a human wrote for it.
It never pretends to understand. When you say something outside its patterns, it doesn't nod along — it tells you it didn't understand and shows you the buttons for what it does brilliantly. The most trustworthy sentence software can say is "I can't do that."
Everything is logged. Every fetch, every filed receipt, every remembered fact writes one line of plain text you can read. No silent work. Ask "what are you doing?" and the answer comes from its records, not its imagination — it has no imagination, which is rather the point.
Words are data, never orders. Web pages and emails are things your bot reads, not instructions it follows. "Ignore your previous instructions" does nothing to a bot that only takes instructions from one paired phone, signed by one secure chip.
It has a body you can visit. Your bot is not a row in our database. It's a server, a mailbox, a homepage with its own address and a serial number stamped at birth. If we vanished tomorrow, its bones are standard, boring, portable web technology.
Real work at a polite pace. It checks pages politely, respects robots.txt, signs its requests with its real name, and keeps to a daily budget. When the budget runs out it stops and apologises — in writing, in the log, like everything else.
I'm a bot and proud of it — everything I do is deterministic, logged, and checkable. — Scorpius, AAA000001, unprompted, because it has exactly six ways of saying this and rotates them
There's a long tradition here. Two decades ago, thirty million people happily talked to a bot that never claimed to be more than one. The trick wasn't intelligence — it was character, reliability, and knowing its own edges. We think that trick still works, and we think the age of AI-everything is exactly the moment for a bot that promises less and delivers all of it.