Cloud AI ruled out for your app's data?
We build a small open model taught only your app's tasks, running privately and offline inside your users' own phones. We are looking for design partners whose software handles data it cannot send to a cloud AI.
Bring a task your app needs to handle and the phones it needs to run on.
Our first build is lab-tested on real phones. No paying customer yet, and no pilot running.
A defined app task and a reason to run it locally
Data that cannot go to a cloud AI
Health, banking and lending, insurance, legal and field-operations software.
Per-question cloud bills
Apps that pay a cloud AI for every question their users ask. A cloud cost comparison is not measured, so we quote no saving.
Not a fit
You need a general chatbot, or your software is cloud web software with no data constraint and no cost pain.
Our first build: offline pharmacy tasks on a real phone
Nesayo is pharmacy software. The live Nesayo product uses cloud AI; its offline edition runs our model inside the phone.
Lab-tested on real phones in Firebase Test Lab (FTL). Test phone: Galaxy A05s, 4 GB, Snapdragon 680. Android 15, airplane mode. FTL
A stock question answered on the phone, in airplane mode.
One sentence turned into a bill.
The model in memory, on a phone with 4 GB in total.
The first turn of that run took 13.4 s, right after a print preview; this remains open. FTL The stock-answer timing is not a first-turn guarantee.
These measurements describe the Nesayo offline edition. Your app's accuracy and device timings still need to be measured. See the measured results and their layers.
A shop pilot is next. Battery use is not measured. iPhone and NPU support are roadmap items, not available.
Define the task. Freeze the exam. Test the device.
A phone timing tells you how a task ran on that phone. It does not tell you whether the model is accurate on your app's tasks. The build process needs both checks.
- Define the work. Map your app's tools, build a sandbox and freeze an exam from real traffic.
- Teach the model. Teach a small open model from a larger teacher with checks, then add code guards.
- Test the build. Quantise it and test it against real-device gates on your users' phones.
This is the proposed process for a partner build. It is not evidence of a deployment or a result on your app.
Questions before a first call
Does model inference send our users' data to a cloud AI?
No. The model runs privately and offline inside your users' own phones, with no frontier API at run time. This describes model inference, not every data flow in your app.
How would we know it is accurate on our app?
We freeze an exam from your app's real traffic and score the model on it. The Nesayo device measurements do not establish accuracy on your app.
Do you have customers already?
No paying customer yet, and no pilot running. Our first build is the offline edition of Nesayo, lab-tested on real phones. A shop pilot is next.
Is it a compressed version of a big model?
No. It is a small open model taught only your app's tasks.
What does it cost?
Commercial terms are not public. Talk to us.
Bring the task your app needs to handle
In the contact form, tell us what your software does, a task you want the model to handle, your target devices and what prevents you from using cloud AI. If per-question cost is the constraint, tell us that.