HoneyPot scores 8.2. Here is the reasoning behind it.

What 8.2 means, and what it does not

Companion apps are easy to rate generously. They all demo well in the first ten minutes, because that is the part every one of them tunes. This score comes from what happens afterwards — session eleven, day nine, the second persona.

A companion app is only as good as its eleventh session.

HoneyPot runs in the browser. You define a character — archetype, tone, pacing, a short list of interests — and then hold a text conversation with it. There is no matching queue, no public profile and nothing to install. The whole product is one screen, which is unusual in a category that keeps adding surfaces.

Persona settings behave like constraints rather than a single opening instruction, and that distinction is what the score rewards. Apps that inject the persona once tend to drift back toward a generic register after twenty or thirty exchanges. The tone we set on day one was still in place on day fourteen, pacing included, which is the harder half to hold.

Recall is selective rather than total. Named details — an occupation, a habit, a running joke — came back unprompted. Broad summaries of earlier sessions did not. In practice that is the right trade, though it means the model occasionally reintroduces something already settled. The free tier keeps less than the paid one, and the gap becomes noticeable in the second week.

The gaps are predictable. Voice and image generation sit on the paid tier, so the free experience is text only. There is no native app, which is a privacy advantage and a convenience cost at once. Content limits are enforced and stated up front. Anyone who wants media volume rather than conversation is at the wrong end of this category.

• Two personas, fourteen days, every session logged
• Recall checked after a deliberate seven-day gap
• Free tier tested before any paid feature
• The same rubric applied to every app compared