What should a personal app remember?
If you tell a meal-planning assistant that you cook for two, you'd reasonably hope not to repeat it every week. That's a useful job for app memory. The harder part is deciding what happens when the information changes, or when an exact dinner plan gets mixed up with a general preference.
A dinner plan and a cooking preference
“Pasta on Monday” belongs in the plan. “Usually cooks for two” is background that may help the assistant suggest next week's dinners. The conversation where you mentioned both is another thing again.
Saved records hold the details you want to look up and change precisely. Chat history lets you revisit what was said. Memory carries selected context into later conversations so you can continue without explaining everything from the beginning.
In Kaddy, the assistant can keep this kind of memory for its own app. It is not a complete transcript, and you should not rely on it to preserve every date, quantity, or passing comment. If Monday's dinner matters, save it in Monday's field.
Then someone comes to stay
Suppose you have a guest for a month. “Cooks for two” was a helpful assumption yesterday; today it produces the wrong portions. You can explain the change in conversation and inspect the app's memory to see what it has retained.
If the old assumption keeps returning, correct the memory directly. Include the temporary nature of the change: “Cooking for three until the end of October.” That is more useful than replacing one undated statement with another you'll have to remember to undo.
This is why editable memory matters in an everyday app. Preferences aren't permanent, and an assistant can misunderstand them. After making a correction, check the next suggestion. The stored context helps the conversation, but it doesn't guarantee the plan is right.
A layout request shouldn't become a preference
When building a meal planner, you might ask for a vegetarian example to test the layout. That doesn't necessarily mean you eat vegetarian meals. Kaddy keeps the build conversation separate from the regular app conversation and uses the latter for memory reflection.
Memory is also separate between apps. Discussing dinner with the meal planner doesn't teach your reading assistant about it. Starting another regular conversation within the same app, however, doesn't mean starting its memory over.
Those boundaries help make sense of a surprising answer. Check which app you're in, what you told it in the regular conversation, and what its memory actually says.
When an answer keeps getting the same thing wrong
Open the app's memory view and look for an outdated or overly broad statement. “Avoids long recipes” might have come from a busy week even though you enjoy cooking on Sundays. A correction such as “Quick dinners on weekdays; longer recipes are welcome at weekends” gives the assistant better context.
Remove details that no longer help. Keep exact commitments in the app's records, where you can see their status, rather than relying on a summary to recall them.
Also check the selected model when deciding what information to share. Kaddy offers models hosted on Cloudflare and models served by external providers. Being able to edit a memory doesn't determine where that context is processed.
Would you otherwise have to explain it again?
That is a useful test for what belongs in memory. A reading assistant may benefit from knowing that you prefer short books during a commute. A project assistant may need the reason you're doing a renovation. Neither needs an ever-growing profile just to help with the task in front of you.
For a simple list, you may use the assistant very little. For a routine you discuss regularly, a few accurate pieces of context can make returning easier. The meal-planner guide shows how to keep the plan itself clear while leaving room for that conversation.