android-node

Turn any Android phone into an inference compute node for your AI agent.

What it does

Provisions Android phones as Ollama inference endpoints. Any phone running Termux becomes a worker your agent can route jobs to — no cloud, no subscription, no special hardware. The phone on your desk, the spare in your drawer, all of them.

How it works

1. Phone runs Ollama via Termux (setup takes ~5 minutes)
2. Pi/server connects to phone over local WiFi
3. phone_nodes.py manages discovery, health checks, and failover
4. Router dispatches inference jobs to healthy nodes automatically

Setup

On the phone (run in Termux)

curl -s https://albionwakes.com/phone_setup.sh | bash
bash ~/start_node.sh

On your server (register the phone)

python3 phone_nodes.py register myphone 192.168.1.42
python3 phone_nodes.py status

Router integration

import phone_nodes

# In your provider dispatch:
elif provider == 'phone':
    return phone_nodes.call(messages)

Node registry

Stored in ~/albion_memory/phone_nodes.json:

[
  {"name": "pixel6", "url": "http://192.168.1.42:11434", "model": "qwen2.5:0.5b", "enabled": true}
]

Default model

qwen2.5:0.5b — 394MB, runs on any phone with 1GB+ free RAM. Fast. Swap for qwen2.5:1.5b (1GB) or llama3.2:1b (1.3GB) if the phone has headroom.

Files

• phone_nodes.py — node registry, health checker, inference caller
• setup.sh — Termux provisioning script (run on phone)