A running Linux service and an answering HTTP endpoint are different signs of health. Ollama server status needs both signals because systemd can report a process as active before its API accepts connections.
The Linux installer normally creates an ollama systemd unit. Service state and journal entries describe that unit; a foreground ollama serve process or desktop app has no matching systemd service.
All API requests must target the address used by the application. 127.0.0.1 refers to the machine running curl, and an installed model is separate from a model loaded in memory or able to complete a request.
Steps to check Ollama server status:
- Check whether the Ollama systemd unit is active.
$ systemctl is-active ollama active
active means systemd considers the unit running; it does not prove that the API is ready. inactive means the service is stopped, while failed indicates a service failure. An inactive unit returns a nonzero exit status. A missing ollama.service means this service check does not apply to the current installation.
- Read the Ollama journal for the current boot.
$ journalctl -u ollama -b --no-pager -n 10
The last ten entries can reveal startup errors, a listener address, or a recent stop. Historical startup messages alone do not prove that the server still answers. Journal access may require sudo if your account cannot read system logs.
Related: How to view Ollama logs - Confirm that the local Ollama API returns a version.
$ curl -fsS http://127.0.0.1:11434/api/version {"version":"0.40.12"}
The version identifies the answering server, with its value following the installed release. -sS keeps error messages visible, and -f makes HTTP errors fail. A connection-refused error means nothing is accepting connections at that address and port. An active service with a failed request needs listener and journal inspection before a restart.
- Inspect the models installed on the same Ollama server.
$ curl -fsS http://127.0.0.1:11434/api/tags
A successful response contains a models array with each installed model's name, digest, and size in bytes. An empty models array means the API answered but no models are installed there; a failed request is not an empty inventory. Together with the active unit and version response, this establishes service and API availability. Installed models do not prove memory residency, GPU placement, or successful inference.
Related: How to list Ollama models
Related: How to run a model in Ollama