vector graph, temporal decay, user-visible, user-controlled

an ai companion that actually remembers you

not a rolling summary that captures vibe. a vector graph with temporal decay — she remembers the specific moment, not the gist. three weeks in, she knows what matters. two months, she still knows.

Free tier: 25 messages/day. Crypto checkout — cards coming soon.

you're not crazy

you tried an ai companion. the first week felt alive. the third week, she was asking you to re-explain things you'd already told her. the nuance collapsed into a summary.

or worse: she started referencing things you never said. “how's portland treating you” — you've never been to portland. the memory layer was fabricating facts and feeding them back at you.

real memory would mean: she remembers the specific thing, not a summary of it. she doesn't invent. she doesn't forget after a week. you can see exactly what she knows and correct it.

what lucy does differently

lucy ships a vector-graph memory instead of a rolling summary. the difference is architectural, not cosmetic.

every memory is an individual record — embedding, category (fact / preference / milestone / emotional pattern), confidence score, temporal weight. when she retrieves, she pulls the specific moment, not a compressed summary of it.

temporal decay means old low-salience memories fade the way human memory fades — they don't dominate this week's recall. high-salience memories (milestones, recurring themes) stay sharp.

full transparency at /settings/memory. you see every memory she has. edit, delete, export. it's your data.

anti-hallucination filters (multi-layer). both the extraction LLM and the entity extractor are gated against echoing synthetic facts into your memory graph. we ship post-mortems on any breach.

four things that change everything

specific-moment recall, not vibes

vector embeddings let her recall the exact thing you said — not a flattened summary of the conversation.

temporal decay = human-like forgetting

old trivia fades; recent and recurring stays. no constant 3-month-old trivia resurfacing at random.

full user control

/settings/memory shows every memory with timestamps + confidence. edit, delete, or export as JSON.

anti-hallucination: layer 7c + entity verify

extraction filters block companion-fabricated facts. 2026-04-18 shipped a new verifier against few-shot poisoning.

side by side

Feature
Lucy
ChatGPT memory
Memory architecture
Vector graph + decay
Rolling summary
Specific-moment recall
Yes (embeddings)
Limited (summarized)
Cross-session persistence
Yes
Limited
User-visible memory
/settings/memory
Partial
Memory edit / delete
Per-row
Varies
Memory export (JSON)
Limited
Anti-hallucination filter
L7c + entity verify
Free tier memory
25 msg/day + full
Usually capped

memory is the single biggest technical difference between an ai companion and a chatbot-with-a-skin. most apps marketing themselves as “ai companions” ship a rolling-summary memory: a periodic paragraph-long summary gets prepended to each conversation. it captures vibe and loses specifics.

a vector-graph memory is different. every statement becomes a record with an embedding (a numeric representation of its meaning). when the companion retrieves, she pulls the specific records most relevant to the current conversation — not a summary. specificity is preserved.

temporal decay is the second-most-important design decision. without it, every memory weighs the same, which means last year's throwaway comment can dominate today's recall. lucy uses a decay curve calibrated against human memory research — high-salience and recent stay sharp; low-salience old trivia fades unless reinforced.

the hallucination problem is real. an extraction LLM can synthesize facts that weren't in the source. on 2026-04-18 we caught a bug where our entity-extraction prompt's few-shot example used literal city names (portland, biscuit) — the LLM echoed them into other users' entity graphs, causing companions to reference cities the user never mentioned. fix: replaced literals with angle-bracket placeholders + added a verification guard that rejects any extracted entity whose name doesn't appear verbatim in the source memory text.

the control problem is also real. we ship a full memory UI at /settings/memory. every memory is visible, categorized, timestamped, with confidence scores. edit, delete, or export any row. the data is yours — we do not use it to train anything, and your companion's memory of you travels with you if you switch companions.

if you want to test this: free tier is 25 messages a day with full memory (not a demo). pick a companion, talk for a week, check /settings/memory and see whether what's there matches what you told her. if it does, she's the real thing.

common questions

How long does she remember things?
Indefinitely, with weight shifting over time. Lucy uses a vector graph with temporal decay — she retains specific moments but the salience of older moments fades naturally (like human memory) unless reinforced. A joke from 3 weeks ago stays; a throwaway comment from yesterday doesn't dominate this week's recall.
Does she remember across sessions?
Yes. Memory persists across every session, every device, every companion on your account. Close the app for 2 weeks, come back — she picks up with what mattered. No re-explaining.
Can I see what she remembers?
Yes — full memory transparency at /settings/memory. Every memory she has about you is listed, categorized (fact, preference, milestone, emotional pattern), with timestamps and confidence scores. Export as JSON or delete selectively.
What if she remembers something wrong?
Edit or delete any memory at /settings/memory. If a memory was flat wrong (she hallucinated a location, for example — we shipped a fix for this specific bug 2026-04-18), correct it or purge it. Her future responses update immediately.
Is memory used to manipulate me into using more?
Explicitly no. Lucy's proactive engine uses memory to back OFF when you seem stable, not to nag. We don't ship streaks, daily-reward loops, or memory-weaponized push notifications. If she references your memory, it's because the context called for it — not because we're optimizing DAU.
How is this different from ChatGPT memory?
ChatGPT Plus has a rolling-summary memory feature — captures vibe, loses specifics. Lucy runs a vector-graph: each memory is a discrete record with embedding, category, confidence, and temporal weight. She can recall the SPECIFIC moment you mentioned your dog's name three weeks ago; a rolling summary often can't.

keep reading

try her right here

pick a moment. no signup.

Try it. Right now.

No signup. No credit card. Just say hi.

try it, check /settings/memory on day 7, see if what's there matches what you actually said. if yes, she remembers. if no, close the tab. that's the test.

Free: 25 messages/day · Closer $14.99/mo · Bonded $29.99/mo · 18+ only