Dawn AI
Companion memory
How Dawn remembers up to 50 key facts about you across conversations, and how to view, edit, or remove specific memories.
Quick answer
Companion memory is what makes Dawn useful past the first conversation. Without it, every session would start from zero. With it, Dawn references your context naturally — your name, your triggers, what's worked before, what you're working on now.
What gets remembered
Memories are extracted automatically from conversations when Dawn identifies something significant. They fall into five buckets:
| Type | Examples |
|---|---|
| Personal context | Name, occupation, family situation, recovery stage |
| Triggers | "Stress at work", "Sunday evenings", "Wine at dinner parties" |
| Coping strategies | "Walking helps", "Calling a friend works better than texting" |
| Active goals | "30 days clean by [date]", "Reduce evening cravings" |
| Preferences | "Be more direct", "Don't suggest exercises unless I ask" |
How memory ranking works
Dawn's memory has a 50-item cap. New memories don't push old ones out by recency — they push by importance.
A new "I tried thought reframing yesterday and it helped me sleep through a craving" carries higher rank than "I had cereal for breakfast", so the cereal goes first if something has to.
The ranking model considers:
- Recency. Recent facts edge out older ones at equal importance.
- Repetition. Facts you've referenced multiple times rank higher.
- Relevance to active goals. Facts tied to current goals stay longer.
- Confidence. Facts you've explicitly confirmed ("yes, that's right") rank higher than implicit ones.
You don't manage this manually — it's automatic. But you can override it (see below).
View and manage memories
Open companion memory
Browse the list
Memories are shown ranked. Each has the source conversation linked, so you can see where it came from.
Edit a memory
Click Edit to fix wording or correct a misremembered fact. Edits boost the memory's confidence rank.
Pin a memory
Pinned memories never get evicted, regardless of rank. Use sparingly — pinning everything defeats the auto-management.
Delete a memory
Removes the fact permanently. Dawn won't reference it again, though it may re-extract a similar fact from future conversations.
When to manually intervene
You usually shouldn't. The auto-management works well. But three cases warrant manual action:
- A memory is wrong. Edit it (ideally) or delete it.
- A memory is private and you don't want Dawn referencing it. Delete it. Dawn won't re-extract something you've explicitly removed within the same session.
- A high-value memory keeps getting evicted. Pin it.
Common pitfalls
Editing a memory to be 'better' than reality
Defeats the purpose. Dawn calibrates against what you've told it. Sanitizing memories produces sanitized advice.
Pinning every memory
The 50-cap then becomes a hard ceiling. New high-importance facts have nowhere to go. Pin only what you can't afford to lose.
Privacy
- Memories are encrypted at rest with the same per-user key as conversation history.
- Included in your data export as
companionMemories[]. - Permanently deleted when you delete your account.
- Never used for AI training.
Next steps
- Goal setting with Dawn — the most productive thing to put into companion memory.
- How conversations work — the system memory sits on top of.
Still need help?
Pick whichever way of getting help works best for you.
Was this article helpful?