Companion is not tied to Ollama. Anything that speaks the OpenAI API will do — llama.cpp's llama-server, LM Studio, vLLM, LocalAI — or OpenAI itself with your own key.

Start your server

Run the server however you normally would. llama-server and LM Studio both default to port 8080; vLLM and others vary, so use whatever address yours prints on startup.

llama-server -m ./models/my-model.gguf --port 8080

Point Companion at it

Open Companion, and in the setup tab enter the endpoint, including the /v1 path:

http://localhost:8080/v1

Then type the name of the model your server is serving. PRSS guesses a default for Ollama and OpenAI, but a self-hosted server names its own model, so there is nothing to guess.

API key, only if you set one

If you started the server with a key — llama-server's --api-key, for example — put it in the API key field. Leave it empty otherwise; a local server usually needs nothing.

Context window

Longer pieces need a bigger context window. Raise or lower it in the same tab to suit your model and your machine.

Start and stop buttons only appear for Ollama running on this machine. PRSS will never try to launch or kill a server it did not start, including a remote Ollama.

Then use it

Select text in the editor and right-click to rephrase, expand or shorten. Try prompts in the Chat tab, or hand Batch Create a list of titles and let it write each post into your blog.