CanIRunAICanIRunAI
Back to blog

low-end-pc

Ollama on Android with Termux: what actually works

A practical walkthrough for testing local AI on Android through Termux, with realistic limits for RAM, battery, and small models.

Kaua Miguel/2026-05-05/1 min read

Android can run local AI, but keep expectations realistic

Modern phones can run small models, especially with 8GB RAM or more. The problem is that memory and battery are shared with the whole system. A model that feels fine on a PC can be slow or unstable on Android.

The safest target is a 1B to 3B model with short prompts.

Install Termux and test

Install Termux from the source recommended by the project, update packages, and add basic tooling:

pkg update && pkg upgrade
pkg install curl

Then follow Ollama's official Linux install when compatible with your environment:

curl -fsSL https://ollama.com/install.sh | sh
ollama serve

In another Termux session:

ollama pull tinyllama
ollama run tinyllama "Explain local AI in one sentence."

Practical tips

Keep the phone charging, close background apps, and avoid multimodal models. If the device gets too hot, stop the test. Mobile local AI is great for learning, but it is not a replacement for a PC with a good GPU.

Read next

Local AI on a low-end PC: 7 settings that actually helpPractical adjustments for running local models on older machines, low RAM systems, or PCs without a dedicated GPU.Can you run local AI without a GPU?A practical guide to CPU-only local AI, model sizes, performance limits, and when a GPU upgrade is worth it.Best local LLMs for 8GB RAM in 2026Lightweight models, quantization choices, and realistic expectations for running local AI on an 8GB RAM PC.