Skip to content

More Than One Machine

One machine can host the models while others use them. The common shape: a desktop with a good GPU does the heavy thinking, and a laptop, a Raspberry Pi, or a phone on the same home network sends it questions through the API. The small machine needs no GPU at all — it just sends text and gets text back.

OLLAMA_HOST, in plain language

By default, Ollama's API listens on localhost — only programs on that same machine can reach it. That's the safe default from The Ollama API.

To let other machines on your home network reach it, you set an environment variable before Ollama starts:

System How to set it
Linux / Mac (terminal) export OLLAMA_HOST=0.0.0.0:11434 then start Ollama from that terminal
Windows setx OLLAMA_HOST "0.0.0.0" in a command prompt, then quit and restart the Ollama app

0.0.0.0 means "listen on all network interfaces, not just localhost." After that, other machines reach the API at:

http://<host-machine's-address>:11434

The host machine's address is its LAN address — something like 192.168.1.50, found in your network settings — not localhost, which on the other machine would mean itself.

Check: from the second machine, open http://<host-address>:11434/api/tags in a browser or with curl. You should see the host's model list. If it refuses, see Troubleshooting.

Which machine does what

Machine Job Needs
The host (e.g., a desktop with a good GPU) Runs Ollama, holds the models, does the math Memory for the model (the size formula still rules)
The clients (laptop, Pi, anything) Send prompts, display answers Just a network connection

A Raspberry Pi makes a fine client: it can't comfortably run an 8B model, but it can ask a desktop to run one all day.

Safety: home network only

Read this twice:

  • OLLAMA_HOST=0.0.0.0 exposes the API to everyone on your network — anyone connected can use your models and read your prompts. On a home network you control, that's usually fine.
  • Do not forward port 11434 on your router. That would expose the API to the entire internet — strangers running prompts on your hardware, on your electricity. There is no login on the Ollama API; the network boundary is the security.
  • When in doubt, leave the default (localhost-only). You lose nothing except remote access.

Local models can still produce wrong or harmful content, and now that content can come from any machine on your network. The privacy win of local AI — your words stay home — only holds if "home" doesn't accidentally include the whole internet.

Next: Troubleshooting