The ButlerUseful technology. Clearly explained.

Home / Useful resources

The Butler practical library

Ollama connection refused: check the listener before reinstalling

Check whether Ollama is listening, confirm the model inventory, and isolate client connection problems before reinstalling.

Reviewed November 14, 2026. Official-documentation-based guidance; not a claim that every installation or hardware combination was hands-on tested.

A refused connection usually means the requested address and port have no service accepting the connection. It is not the same as an incorrect model name or an exhausted API balance. First establish whether Ollama itself answers; only then troubleshoot the app calling it.

Three checks, in order

  1. Open Ollama from your applications or Start menu. If you installed only the command-line version and no server is running, use ollama serve in a terminal. Leave that terminal open. If it reports that the address is already in use, do not keep starting copies: a service may already own the port.
  2. On Windows PowerShell, run Invoke-RestMethod http://localhost:11434/api/tags. On macOS/Linux, run curl http://localhost:11434/api/tags. A JSON response containing a models array means this API is reachable from that machine. An empty array means the connection works but there are no listed local models.
  3. Run ollama list and select an exact installed model name in your client. If the API works here but your client fails, compare its configured server address. Inside a container, localhost refers to that container, not automatically your computer.

If it still fails

Read the server log rather than changing firewall rules blindly. On Windows, open %LOCALAPPDATA%\Ollama\server.log in a text editor. On a systemd Linux installation, journalctl -u ollama --no-pager --lines=80 shows recent entries. On macOS, check ~/.ollama/logs/server.log.

Do not expose port 11434 publicly to fix a local connection. If you ask for help, include your OS, Ollama version, installation method, the attempted address and the relevant error lines. Remove prompts, usernames, keys and other private content from logs first.

Verify the fix

Repeat the API check, then send a short harmless prompt through the original client. A successful API listing alone does not prove model inference works.

Documentation checked November 14, 2026. These are diagnostic steps, not a claim that every platform was installed and tested by The Butler.

Official references