HoneyPot Memory FAQ

Concrete, durable statements are the ones that survive: your name, what you do, who matters to you, preferences you state clearly, boundaries you set. Passing mood and small talk generally do not become entries, which is deliberate — a store full of what you ate on Tuesday retrieves worse than a small store of things that stay true. The practical rule is that anything you want remembered should be said outright at least once, rather than hinted at.

Because saving and recalling are two separate steps and either can miss. A detail has to be captured as an entry first, which favours explicit statements over implications. Then, on the turn she replies, retrieval has to judge that entry relevant to what you just asked. A perfectly saved fact can sit unused if nothing in the current message points at it, which is why a direct question usually surfaces it immediately.

There is a practical limit rather than a hard wall, and it moves with your tier and how tidy your store is. Retrieval works by relevance, not by scrolling backwards, so an entry from months ago can surface faster than something from last week if it fits better. That is also why pruning matters: crowding the store with near-duplicates makes the older, sharper entries harder to reach.

Open the memory list. Entries are shown individually rather than as an opaque summary, which is the whole point of storing them separately from the transcript. Most people find two or three misreadings on their first look, usually a detail attached to the wrong person or a preference recorded from a joke. Fixing those is a couple of minutes of work and changes the tone of the next conversation immediately.

Deleting removes it from what gets assembled into future prompts, so she stops bringing it up from the next message onward. What deletion cannot do is retroactively edit earlier messages in your history, since those are the record of what was actually said. If a subject is genuinely off limits, the durable fix is to delete the entry and add the boundary to her persona layer, so it is re-stated on every turn instead of depending on retrieval.

Yes, and it is the cleaner approach. Set the off-limits subjects during persona setup rather than correcting them later. Boundaries placed in the persona layer are present on every single turn, which makes them far more reliable than an instruction dropped mid-conversation and hoping it gets retrieved. You can add to that list at any point.

Three habits, in order of impact. Prune stale entries, because a clean store retrieves more precisely than a large one. Pin the handful of facts that define the relationship, so they are never competing for space with trivia. And open long sessions with a one-line recap of where you left off, which hands retrieval an obvious target. None of that is glamorous, and all of it works better than repeatedly re-stating the same detail.

That is usually a symptom of too much memory rather than too little. When the store is crowded, more entries clear the relevance bar on any given turn, and the reply fills up with references instead of responses. Trim the entries that no longer matter, unpin anything you pinned reflexively, and she will start engaging with the message in front of her again.

No. Each companion keeps her own persona sheet and her own fact store, so a new character genuinely starts new — which most people want, since a second companion inheriting the first one's inside jokes would be unsettling. If you do want a head start, restate the facts about yourself during her setup and you skip the first week of groundwork.