Can your app's AI work on a 4 GB phone without a signal?
In our first build, a small open model taught only one app's tasks answered a stock question offline in 4.25 s. The first turn of that run took 13.4 s (right after a print preview); still open. FTL
Whole-app test conditions: Galaxy A05s, 4 GB, Snapdragon 680. Android 15, airplane mode, run PASSED. Test build, 31 Oct; no later merged build has passed a whole-app run yet. Lab-tested on a real phone in Firebase Test Lab (FTL). See the measured results and limits.
Our first build is Nesayo (pharmacy software, nesayo.com). No customer uses the offline model yet; a shop pilot is next.
Which decision are you making?
For software teams whose data can't go to a cloud AI. Choose the guide closest to your question.
Can a small language model run on a budget Android phone with 4 GB RAM?
Memory, answer time and the separate speed benchmark. Start here to assess what the tested phone tells you, and what you still need to measure on your users' devices.
Read the device guide Offline readinessCan an AI assistant in a field app work offline, with no signal?
What must be on the phone before signal drops, and what fell short offline in our first build. Includes a loan collection illustration: not built, not measured.
Read the offline guideRead the evidence by layer
- FTL: lab-tested on a real phone in Firebase Test Lab. Check the conditions for each run.
- EVAL: scored on a frozen test on a server.
- SIMULATED: synthetic shops, offline build on the VM; not device proof.
For the speed benchmark, network was not controlled (not airplane mode). Airplane mode belongs to the whole-app run above. Read the raw llama-bench rows and run conditions.
What would you need to test in your app?
Tell us what your app does, one task you want to run on the phone, and the phones your users carry.