Companion

The text becomes typed questions instead of a completion. Each turn is a single call that answers nine questions at once, and code decides from those answers whether to act, ask, or shrug. Nothing on the page describes what happens to your input beyond that, and no author or repository is named.

It answers a fixed set of slots rather than whatever you meant, and you get a shrug on anything outside them, which the sample inputs show deliberately, including an insult. The trade is a predictable, cheap turn with no free-form understanding behind it.

Every turn is labeled with the question count and single call, and the interface shows what the robot decided along with the reason. Mood and focus are tracked too, from neutral to idle or near bed. That exposes the branch the code took, though not the underlying scores that produced it.

The page addresses none of that. No key, latency, or price detail is published, and it presents as a hosted demo with sample prompts you can click. Assume calls are billed per turn, and check the key and rate situation yourself before taking it past the built-in examples.

The page offers clickable samples such as go to the lamp, come here quick, hide, can you dance for me, look at the window, take a nap and is anyone else home. Each turn is a single call answering nine questions at once, and code then decides whether to act, ask or just react.

It shows what the robot decided and why, plus a mood that starts neutral and a focus value such as idle or near bed. Every turn is labelled with its question count and single call.

No. The page names no author or repository and publishes no latency, key or price detail, presenting as a hosted demo with sample prompts. Assume each turn is billed, and verify the key situation yourself before taking it past the built-in examples.