Use Ollama running on another computer, such as a Mac or a Linux box

NeuralVault runs on Windows only, but the models can run on any machine on your network that runs Ollama. Make that machine listen on the network, then point NeuralVault’s Ollama address at it in one small file.

Last checked against the app on · View as Markdown

Steps

  1. Be clear which machine does what. NeuralVault itself is a Windows app: there is no Mac or Linux version. The other machine, a Mac or a Linux box or another Windows PC, only runs Ollama and holds the models. Install Ollama there and pull a model, for example "ollama pull llama3.1".
  2. Make Ollama listen on the network. By default it binds 127.0.0.1 port 11434, which only that machine can reach. Ollama’s documentation says to change the bind address with the OLLAMA_HOST environment variable.
  3. On a Mac running the Ollama app, run launchctl setenv OLLAMA_HOST "0.0.0.0:11434" in Terminal, then restart the Ollama app.
  4. On Linux running Ollama as a service, run "systemctl edit ollama.service", add under [Service] the line Environment="OLLAMA_HOST=0.0.0.0:11434", save, then run "systemctl daemon-reload" and "systemctl restart ollama".
  5. On Windows, quit Ollama from the taskbar, open Settings and search for environment variables, choose "Edit environment variables for your account", create OLLAMA_HOST with the value 0.0.0.0:11434, press OK, and start Ollama again from the Start menu.
  6. Find that machine’s address on your network, for example 192.168.1.50. From the Windows PC, open http://192.168.1.50:11434/api/tags in a browser, using your address. A page of JSON that lists your models means the machine is reachable. NeuralVault asks the same address for its model list.
  7. In File Explorer, type %USERPROFILE% in the address bar. Open the file .neuralvault-ollama.json in that folder, or create it, and make it contain one line: {"base":"http://192.168.1.50:11434","model":"llama3.1"}. Use your address and a model you pulled. Keep the quotes, and no space before a colon. Save it.
  8. Open Settings, then Intelligence, then API keys. The Ollama row now says it is reached at your address and shows its status.
  9. Open the model button in Chat and choose a model from that machine. Choosing a model keeps the address you set.

How to tell it worked

The Ollama row reads "Reached at" followed by the other machine’s address with the status "Ollama is answering", and a message in Chat is answered.

If it does not work

The browser test cannot reach the address.
Ollama is still bound to its own machine, or was not restarted after you set OLLAMA_HOST, or something between the two computers is blocking the port. Check the setting and the restart first. A firewall on either machine is a common cause, which is general networking and not something Ollama’s documentation covers.
The Ollama row still says localhost.
The file is not being read as written. It must be in your user folder, named .neuralvault-ollama.json, and hold both "base" and "model" in quotes with no space before the colon. A file that does not parse falls back to localhost without a message.
The status says Ollama is not running, or has no models installed.
It is reporting on the address you set. Pull a model on the other machine with "ollama pull" and press recheck.
Settings says it writes the address for you.
Some messages in this build still say so. Settings shows the address but has no box to change it, which is why the file edit above is the way.