SideNote Pro

Use Ollama with SideNote Pro for local AI

Ollama runs language models on your own hardware and exposes an OpenAI-compatible interface. SideNote Pro can act as the Windows client for that endpoint: point it at Ollama and your prompts go to localhost rather than to a hosted provider, so inference stays on your machine.

What Ollama is, and what it is not

Ollama is separate software that downloads and runs language models locally. SideNote Pro does not bundle it, does not install it, and does not supply the models. What SideNote Pro provides is a client that can talk to it.

Ollama exposes an OpenAI-compatible API alongside its own, which is what makes this work. SideNote Pro treats it as any other compatible endpoint: GET /models to test and discover, POST /chat/completions to generate.

Setting it up

  1. Install Ollama and pull a modelFollow Ollama's own documentation. You need at least one model downloaded before anything will answer, and the model name you pulled is the model ID you will use.
  2. Make sure the Ollama server is runningThe client and the server are separate. If SideNote Pro cannot reach the endpoint, this is the first thing to check.
  3. Add an Ollama profile in SideNote ProSettings, AI providers, add a profile, and choose Ollama from the Provider list. The local address Ollama normally serves on is filled in for you.
  4. Leave the API key emptyA default local Ollama install does not require authentication. If you have put it behind a gateway that does, enter that gateway's key.
  5. Select Test & Load models, then pick oneIt asks your local server what it is serving and fills the Model picker with the models you have pulled. Choose one, or type its name.
Ollama profile fields
FieldValue
ProviderOllama
Base URLFilled in for you: http://localhost:11434/v1
ModelThe model you pulled, by its Ollama name
API keyEmpty for a default local install
Reasoning effortProvider default
Context windowAuto

Context window in SideNote Pro does not change the context Ollama runs a model with. Ollama uses 4,096 tokens by default, unless OLLAMA_CONTEXT_LENGTH or the server's own configuration sets another size, and an Ollama profile shows a hint saying so below the setting. If your server's size differs from Auto's 128K, choose Custom and enter the size the server actually uses, so SideNote Pro plans each request to fit it.

The address arrives with the /v1 suffix already on it, which is the part that matters: http://localhost:11434 is Ollama's own API root, and http://localhost:11434/v1 is the OpenAI-compatible one. Edit the field if your Ollama runs on another machine or a non-default port. It is an ordinary text box, not a fixed value.

Settings › AI providers

The profile strip at the top of AI providers, with one profile marked Default and another beside it, above the Add profile, Duplicate and Delete buttons.
A local profile does not have to replace a hosted one. Keep both, mark whichever you use most as the default, and switch per conversation from the picker in the composer.

What actually changes when you go local

The good part

Requests go to localhost. Inference happens on your hardware, with no hosted API charge. Rate limits and whether anything reaches a hosted provider depend on how you configure Ollama.

The cost

Speed and capability now depend on your machine. Local models that run comfortably on a laptop are generally smaller and less capable than the best hosted models, and long prompts are noticeably slower without a capable GPU.

SideNote Pro cannot make guarantees about third-party software. What it controls is the destination of the request. Whether that destination keeps everything on your machine depends on how you configured Ollama, and Ollama's own documentation is the authority on that.

Setting expectations for local models

  • Image input is uncommon. Most small local models are text-only. If images seem to be ignored, the model is the reason. See screenshots and images.
  • `reasoning_effort` may be ignored. Leave it on Provider default unless the specific model documents support.
  • Usage metadata may be absent. SideNote Pro falls back to local token estimates. There is no hosted API bill to reconcile for a local model.
  • Long context is expensive in time. A long conversation, a large attached document or a folder-retrieval request means a lot of tokens to process locally. Setting Context window to a smaller Custom value on this profile helps.
  • Context compaction still calls the model. A local model has no API charge, but compaction uses your hardware and time.

You can keep a hosted profile for hard work and a local profile for confidential work, then switch between them in the composer. See using local models for private workflows.

Keep reading