Compact models and quantization
Choosing and compressing small language models so they fit in a phone's memory and storage without losing the skills each app needs.
ON-DEVICE AI RESEARCH
Language models that run on the phone in your pocket
LeapGenix researches how to make local, on-device language models genuinely useful in everyday mobile apps: private by default, available offline and fast enough to feel natural.
WHY LOCAL
Why run AI on the device?
Most AI assistants send every request to a remote server. That works, but it means personal information leaves the phone, help disappears without a connection, and every answer waits on the network.
For the apps we are building — a child's learning companion, a personal planning assistant and a fitness coach — those trade-offs matter. Running the model locally keeps sensitive context on the device, keeps working offline and removes the round-trip to a server.
The hard part is making a model small and efficient enough to run well on everyday phones while still being genuinely helpful. That is the focus of our research.
RESEARCH AREAS
What we are working on.
Efficient mobile inference
Running models on phone CPUs, GPUs and neural accelerators with careful attention to response speed, battery life and heat.
Private personalization
Adapting to each person's progress and preferences using context that stays on the device instead of a remote profile.
Safety and evaluation
Age-appropriate guardrails for children, clear boundaries for fitness guidance, and testing on the kinds of phones people actually own.
IN OUR APPS
One local model layer,
three experiences.
Each LeapGenix app is designed to use the same on-device foundation, tuned for its audience: gentle and age-appropriate for children, practical for personal planning, and encouraging but careful for fitness.
ON-DEVICE LLM
★
Little Champion · Explains ideas at a child's level
In development↗
ForgeLift · Turns goals into weekly actions
In development✦
AIFitCoach · Answers training questions offline
In developmentDESIGN GOALS
What on-device AI should feel like.
- ✦ Private by default
- ✦ Works offline
- ✦ Instant responses
- ✦ Runs on everyday phones
RESEARCH STATUSEarly research.
Early research.
Honest results.
This research is ongoing. We have not published model details or benchmarks yet, and we will not claim results we cannot show.
As experiments mature, we plan to share what we learn, including what did not work. If you work on local or on-device language models and would like to compare notes, we would love to hear from you.
Talk to us about the research ↗