Only the AI your software needs, inside the phone.
First build: an offline stock answer in 4.25 s on a Galaxy A05s, 4 GB, Snapdragon 680 (FTL). The first turn of that run took 13.4 s.
Neykosa builds a small open model taught only your app's tasks, running privately and offline inside the app.
Our first build is for Nesayo, pharmacy software. The live Nesayo product uses cloud AI. Its offline edition runs our model inside the phone and is lab-tested on real phones in Firebase Test Lab (FTL). A shop pilot is next.
The following results are from one FTL run of a test build (build of 31 Oct; no later merged build has passed a whole-app run yet) on the Galaxy A05s, with Android 15 and airplane mode:
| Measurement | Result | Layer |
|---|---|---|
| Stock answer, offline | 4.25 s | FTL |
| Bill from a sentence | 5.76 s | FTL |
| Model memory | 673 MB | FTL |
Source run: ftl_final_whole_app_a05s_26efc64a1_20261031.json.
The first turn took 13.4 s, right after a print preview; that remains open. Read the answer times as individual measured turns, rather than a latency guarantee.
Map the app's tools, build a sandbox, and freeze an exam from real traffic. Teach a small open model from a larger teacher with checks. Add code guards, quantise, and pass real-device gates.
Public measurements, evaluation tools and examples for small on-device models, using public or synthetic data only. Measurements will identify their layer and source run.
We are looking for design partners whose software handles data it can't send to a cloud AI. We build the small model for your app's tasks and test it on your users' real phones.
Talk to Team Neykosa about your app.
Team Neykosa