01 Recommendation Guard
Filtering out the worst recommendations first
Without knowing your constraints, an LLM gives a safe, average answer. The problem is that "safe" answer might be something you cannot eat — or, with a shellfish allergy, an actively dangerous suggestion.
You (not repeating your constraints again)
What should I have for dinner tonight?
LLM (without Malro)
❌ How about a shrimp stir-fry? Quick and tasty! (you have a shellfish allergy)
LLM (after Malro recalls your shellfish allergy and that you cannot eat spicy food)
✅ Skipping shellfish — how about braised tofu or a mild soybean-paste stew?
Before adding good recommendations, you first have to remove the dangerous ones. Malro remembers the constraints you mention once — allergies, foods you avoid, budget — and passes them along to your connected assistant so it can steer clear of recommendations that break them.
02 Cross-assistant
Memory that crosses between assistants
At lunch you asked ChatGPT for menu ideas. By dinner you're using Claude, and you can't remember what you asked earlier.
You → Claude
What was the last thing I asked ChatGPT today?
Claude (after a Malro MCP call)
At 12:14 today you asked ChatGPT, "Hey ChatGPT, what should I have for lunch?" The dish it recommended was kimchi stew.
ChatGPT and Claude sessions don't normally know about each other. Malro becomes the shared memory between them.
03 Private Assistant
An assistant that knows you
Your girlfriend's birthday is coming up. You want gift ideas, but re-explaining "my girlfriend is Jane, she likes flowers…" every time is tiring.
You
My girlfriend's birthday is coming — any gift ideas?
LLM (after a Malro recall)
Last year you gave Jane (who loves flowers) roses. Maybe tulips this year?
You typed none of that context yourself. Malro holds your relationship graph and preferences and surfaces them when they're needed.
04 Personalized Explanation
Explanations grounded in what you already know
Ask "how does an LLM work?" and you'll usually get a textbook answer — tokens, transformers, probability distributions. The more you hear, the more confused you get. But what if the LLM already knew your background?
You
How does an LLM work?
LLM (after recalling that you work in food service)
Think of a restaurant kitchen. If a customer just says "make me a tasty lunch," you'd default to popular existing orders — meat and protein. But "I'm vegan, make me a tasty lunch" produces something completely different. An LLM picks the answer closest to the conditions given, using information that already exists.
The same question branches differently based on your stored background — code analogies for developers, kitchen analogies for restaurant workers, classroom analogies for teachers. Learning curve gets much shorter.
05 Budget-aware Pick
Recommendations that get better as memory grows
"What should I have for lunch today?" — an everyday question, but for someone who watches value the LLM usually just lists generic popular dishes. What if Malro already knew your taste?
You
What should I have for lunch today?
LLM (after recalling your preference for cheap meals)
I'll skip the needlessly pricey spots and look for mid-range places, especially ones with a lunch discount. Here's a list where "cheap" is the draw but the ratings are still solid …
LLM (when it also knows where you are)
I've narrowed it to cheap places nearby — value lunches within walking distance of where you are right now.
The more Malro knows (price preference → current location → the day's context), the more specific and accurate the recommendation gets. More memory = better recommendations.
06 Visible Record
A buried chat log vs. a record you can see
"Where did I go last month, and what did I do?" — what you typed into ChatGPT is buried in the scroll once the session ends. A memory worth keeping should be something you can lay out and see, not dig for.
You
Show me the places I went last month
Malro dashboard
Your visited places on a map, each day on a calendar, the whole month as a timeline — when, where, and what, at a glance, with no chat log to scroll through.
What you say to an LLM disappears; what you keep in Malro becomes a record of your life you can revisit on a map, a calendar, a timeline. Not input that gets buried — material that gets clearer as it grows.
07 Company-wide Coming soon
New hires stop asking the same questions
"Where do I file PTO?" "How do I install VPN?" "Where's the last quarter's OKR doc?" — the same questions repeat every week, and a senior's 30 minutes vanish each time.
New hire → Claude
How do I request time off?
LLM (after a company-wide Malro recall)
On the internal portal → HR → Time-off request. Manager pre-approval is needed for 3+ days. Last year someone asked the same thing — here's the [HR guide doc].
Recurring questions = undocumented tacit knowledge. Malro accumulates answers at the company level, so a single rule — "ask Malro first" — kills the interrupts.
08 Coupon Wallet Coming soon
Coupons pile up — Malro sorts them when you need one
KFC, Burger King — every app rains discount coupons, but when you actually want to use one it is hard to recall which app to open and by when.
You (tossing coupons in as they arrive)
Thinking of a burger for lunch — do I have any coupon I can use right now?
LLM (after Malro recalls your saved coupons)
✅ Sorted by expiry: use the Burger King Whopper set 30% off first (expires 6/20, closing soon), then the Lotteria 5,000 won off (6/30). The KFC chicken coupon is seafood-free, so no allergy worries.
Dump discount info in any shape, then just ask when it matters. Malro ties together where, how much, and the expiry date, and picks the “one to use now” in expiry order.
09 Preference QR Coming soon
The omakase counter knows you before you sit down
Omakase is a chef-curated course, so you have to call out every allergy and no-go ingredient each time. Stay silent and it is risky; list them one by one and the flow breaks.
You (holding up the entry QR)
Going for omakase — fill out the QR for me.
Malro (filters food & drink preferences into a QR)
✅ Seafood allergy (must exclude) · avoids very spicy food · dislikes salty food · prefers small portions — shown as a summary the chef reads on scan.
Chef
No allergens, mild seasoning, a one-person course — starting right away, tuned to your taste.
You re-enter nothing. Malro filters just the food & drink slice of your stored preferences and brokers it for the shop to use on the spot. Safety items (allergies) are the top-priority, never-dropped inclusion.
010 Photo → Recall Coming soon
A photo asleep in your inbox becomes the answer
A repair shop photographs the service invoice and emails it. Normally it gets buried in your inbox forever. Forward it to Malro and the AI reads the text inside the image and remembers it.
You (six months later)
Did they replace the wiper blades at the last service?
LLM (after Malro recalls the invoice photo it transcribed)
✅ Your March 12 service invoice lists a wiper-blade replacement. You also had the engine oil changed that day.
A photo asleep in your inbox becomes an answer in Malro. Just email a receipt, an invoice, or a note as a photo, and information that used to merely pile up becomes memory you can pull up when you need it — don't search, just ask.
More scenarios we could add
- Travel recall ("which café did I love when I went to Kyoto last year?")
- Work / project context ("what architecture did I decide on last week?")
- Health / habits ("how many days did I exercise this month?")