Ollama

Ollama serves models over an OpenAI-compatible API on port 11434.

Start Ollama

ollama serve
ollama pull llama3.2

ollama serve runs in the foreground. On macOS and Windows the Ollama app already runs it for you.

Add the upstream

Ollama usually needs no key. Replace the model ids with the ones you pulled:

noket upstream add \
  --name ollama \
  --protocol openai \
  --url http://127.0.0.1:11434 \
  --model llama3.2 \
  --model qwen2.5

Both http://127.0.0.1:11434 and http://127.0.0.1:11434/v1 work as the base URL; Noket does not duplicate the /v1 when it probes for models.

Then:

noket upstream test ollama
noket up

Model ids

Use the same names you would pass to ollama run, including any tag: llama3.2, qwen2.5:7b, nomic-embed-text. Ollama reports these names in its model list, so the picker in noket upstream add lists them directly.

If Ollama runs on another machine

Point the base URL at that machine, for example http://192.168.1.20:11434, and make sure Ollama listens on more than loopback:

OLLAMA_HOST=0.0.0.0:11434 ollama serve

Only do this on a network you trust. Ollama has no authentication by default, so anything that can reach the port can use the models.

If you would rather not expose Ollama directly, run Noket on the machine that hosts Ollama and keep the base URL on loopback.

Start Ollama

Add the upstream

Model ids

If Ollama runs on another machine

Next