AI can write our emails, plan trips, recommend what to watch and tell us what to do when we're bored.
And somehow, we still end up staring at the same screen.
So I wanted to build something backwards.
Instead of an AI that keeps you online I wanted one that gives you a reason to put your phone down.
Meet Memory Hunt.
The Idea
Memory Hunt takes a personal memory and turns it into a small real world mission.
For example:
“When I was younger, I used to play badminton with my cousins on our terrace.”
Memory Hunt might turn that into:
🌤️ Memory Hunt #01
Go somewhere you used to play as a kid.
Stay there for five minutes.
Find one thing that looks exactly the same.
Find one thing that has completely changed.
Don't take a photo.
Just remember it.
Then the app's job is done.
Close it. Go outside.
Why Memories?
I could have built another app that says:
“Go for a walk! 🚶”
But that doesn't answer the interesting question:
Why would I want to go?
A memory can provide that reason.
Maybe you used to:
- Play cricket with your friends
- Ride your bicycle around the neighborhood
- Visit your grandparents' garden
- Get snacks from the same corner shop
- Sit outside after school
Memory Hunt turns those memories into tiny invitations to revisit the physical world.
Not to recreate the past.
Just to notice it again.
How It Works
The flow is deliberately small:
Remember something
↓
Memory Hunt
↓
Local AI
↓
Outdoor mission
↓
Put phone away
↓
🌱
Go outside
The user provides:
- A memory
- Time available
- Activity level
- Alone or with friends
- Desired mood
The model turns that context into one short mission.
For example:
Memory
“I used to play cricket in the street with my neighbors.”
Mission
🏏 The Old Playground Quest
Go back to that street.
Find the place where you usually played.
Find one thing that wasn't there when you were a kid.
Then find one thing that still feels familiar.
No giant itinerary.
No endless recommendations.
Just a reason to go.
Under the Hood
Memory Hunt uses an open-weight AI model through local inference rather than relying on a closed cloud API for the core generation.
The request is structured before reaching the model:
Memory:
"I used to play cricket with my friends
in the street after school."
Time:
15 minutes
Activity:
Easy
People:
Alone
Goal:
Nostalgic / reflective
The model generates a short mission based on that context.
The mission is then checked against the basic constraints:
- Fits the available time
- Is practical outdoors
- Requires minimal or no spending
- Includes a concrete action or observation
- Doesn't depend on information the user never provided
The final result is intentionally small.
Memory
↓
Structured prompt
↓
Open-weight model
↓
Mission
↓
Go outside
↓
Optional reflection
The user reads it, locks their phone and goes.
Why Open Source Matters
This project deals with something personal:
your memories.
A memory might be:
“My grandfather used to take me to this park.”
I don't want that automatically sent to a remote AI service just to generate a five line activity.
That's where open AI matters.
Privacy
With local inference, memory processing can happen on the user's device, keeping personal memories out of a third-party AI API.
Offline Potential
An outdoor app shouldn't have to depend entirely on an internet connection.
Local inference makes offline generation possible, which is especially useful for parks, trails, campuses and other places with unreliable connectivity.
Freedom to Experiment
The model can be replaced.
That means I can experiment with different open models, prompts, constraints and behaviors without building the entire experience around one closed provider.
Memory Hunt doesn't need a general-purpose chatbot.
It needs one specific capability:
Turn something personal into a small reason to experience the real world.
The Design Rule
I gave Memory Hunt one rule:
The screen should be the shortest part of the experience.
Most apps want more engagement.
More clicks.
More sessions.
More time inside the app.
Memory Hunt is different.
If the result looks like this:
App usage: 20 seconds
Time outside: 35 minutes
that's a win.
The AI did its job by becoming irrelevant.
What's Next?
I'd like to explore:
- Childhood Mode — missions from childhood memories
- Friend Mode — missions based on shared memories
- Nature Mode — memory-inspired outdoor observations
- Place Mode — missions connected to remembered locations
- Surprise Mode — unexpected missions from old memories
The goal isn't to perfectly recreate the past.
It's to use the past as a reason to pay attention to the present.
Why Build This?
We're building increasingly powerful technology to help us do more without leaving our screens.
I wanted to try the opposite.
What if AI could be useful because it eventually makes itself unnecessary?
Memory Hunt doesn't want you to spend an hour talking to an AI.
It wants you to read one small prompt and think:
“Wait... I haven't been there in years.”
Then you close the app.
You go outside.
You look around.
Maybe you remember something.
Maybe you notice something completely new.
And for a little while, the AI isn't the most interesting thing in your life.
That's what I wanted to build for Touch Grass. 🌱
Now I Want to Know Yours
Everyone has a place, person or tiny moment they haven't thought about in years.
Maybe it's the street where you played.
A park you used to visit.
A shop where you bought snacks after school.
Or just a random corner where something memorable happened.
If an AI could turn one of your memories into a real-world mission, what memory would you give it?
Drop it in the comments. 👇
I might turn some of the most interesting ones into Memory Hunt quests.
And maybe that's the whole point:
Don't just tell AI about a memory.
Give yourself a reason to go make a new one. 🌱
Top comments (0)