Error messages

Error: could not connect to ollama server, run 'ollama serve' to start it

Error: could not connect to ollama server, run 'ollama serve' to start it

You type ollama run or ollama pull and get this straight back:

Error: could not connect to ollama server, run 'ollama serve' to start it

Ollama is two programs in one binary. ollama serve is the server that loads models and does the work, and every other ollama command is a client that talks to it over HTTP, at 127.0.0.1 port 11434 unless the OLLAMA_HOST variable says otherwise. Before running a command, the client checks the server is there. We read that check in Ollama 0.34.4: if the connection is refused, the client tries to start the server, and this line is what it prints when it can't.

The wording changed twice

Most search results and old issues quote a different line. We read the same function in older releases:

• up to 0.7.0: Error: could not connect to ollama app, is it running?
• 0.7.1 (May 2025) to 0.15.x: Error: ollama server not responding - followed by the reason, which on Linux was the current message
• 0.16.0 (February 2026) onward: the reason on its own, as above

They all mean the same thing. As of September 26, 2026, 42 issues in the ollama/ollama repo quote the "ollama app" version and 84 contain "ollama server not responding".

Why it can't start the server for you

What happens next depends on the system. On Linux the client never tries: it prints the message at once. On macOS it starts Ollama.app only if the ollama command is a link into that app. A Homebrew or hand-built binary gets the message instead. On Windows it launches ollama app.exe and waits, so there you're more likely to see could not locate ollama app or timed out waiting for server to start.

What to do

On Linux with the standard install, the server runs as a systemd service. Ollama's Linux docs start it and check it with:

sudo systemctl start ollama
sudo systemctl status ollama

Without the service, run ollama serve in a second terminal and leave it open. On macOS or Windows, open the Ollama app, then try the command again.

If the server is running and you still get this, the client is looking in the wrong place. Check OLLAMA_HOST in the shell where you run ollama. It has to match where the server listens, and a server bound to another machine or port won't be found at the default address. A contributor put it simply in #2132: connection refused means nothing is listening on that address and port.

Proxies catch people out too. In #7697 (November 2024) the user had HTTP_PROXY pointing at a local proxy, and the proxy's own address was refusing connections. The collaborator's fix was to set NO_PROXY=127.0.0.1 in the client's environment and leave the server's proxy settings alone, since the server still needs its proxy to download models.

On Windows, the same collaborator hit this after an automatic upgrade from 0.3.14 to 0.4.2 left the install broken, and reinstalling from ollama.com fixed it.