SideNote Pro

Set up any AI provider, listed or not

Connecting a provider used to mean copying an endpoint out of somebody's documentation and hoping you had the right number of path segments. Now you pick the provider and it is filled in. The part worth knowing is what happens when your setup is not the ordinary one.

Let the provider choice do the typing

In Settings, AI providers, the first control on a profile is Provider. Choose OpenAI, Anthropic, Google Gemini, DeepSeek, Ollama, LM Studio, or Custom OpenAI-compatible, and the endpoint that service normally answers on is filled in for you. Add a key if the provider needs one, select Test & Load models, pick a model, and you are finished.

Settings › AI providers

The SideNote Pro provider menu, listing OpenAI, Anthropic, Google Gemini, DeepSeek, Ollama, LM Studio and Custom OpenAI-compatible.
Six services SideNote Pro already knows an address for, and one entry for everything else. Local providers are in the same list as hosted ones, because as far as a profile is concerned they are the same kind of thing.

The getting started guide walks the whole path once, and the OpenAI walkthrough is the same five steps with the reasoning behind each one.

The endpoint is still yours

A provider choice is a starting point, not a mode. Base URL stays editable on every profile, so the same setup works whether you are calling the provider's public API, a company gateway, a reverse proxy, an alternate host, a self-hosted server, or a local model on another machine on your network.

It follows that the list is not a limit. If what you want is not in it, choose Custom OpenAI-compatible, enter the base URL and the model ID, add credentials if the service wants them, and the profile behaves like any other. Configuring any OpenAI-compatible provider has the exact contract to check a service against.

An address you typed yourself will not be replaced behind your back. SideNote Pro fills Base URL in only when the field is empty or still holds one of its own defaults, so switching provider on a profile that carries your endpoint keeps your endpoint, says so, and offers Use the default endpoint for the times you meant the standard one.

You never have to wait for a model list

Test & Load models does two useful things: it proves SideNote Pro can reach the endpoint, and it fills the Model picker with what the provider offers. It is worth running. It is not a requirement.

Model is an editable combo box. Open the list, type to filter it, or type a model ID that was never in a list at all. The placeholder says as much: Enter or search a model ID. A hand-typed ID works whether discovery succeeded, came back empty, is not supported by that service, or was never run.

So a provider without a model list is an ordinary provider here, not a broken one. Copy the exact ID from its documentation, watch the capitalisation and any vendor prefix, and send a message.

One provider can have several profiles

A profile is an endpoint, a model and a few settings, not an account. Keeping four of them against the same provider is normal: a fast model for everyday work, a deeper one for the problems that need it, one pointed at a work gateway, one at your own endpoint.

Duplicate copies the key and the endpoint, so the second profile is a rename and a model change away. Name them for the job rather than the model, and the composer picker still reads clearly in six months. Setting up a model for each kind of task is the version of this with the three profiles worth having.

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.
The strip along the top holds every profile, with the default marked. Switching between them is a pick in the composer rather than a trip back into Settings.

Leave the context window on Auto

Context window lives on the profile, because a small local model and a current hosted one are not remotely comparable on this. Auto is the right setting for almost everyone: it uses the size the provider reports for that exact model where there is one, a size SideNote Pro already knows where there is not, and otherwise the fallback window, which it names rather than hides.

Settings › AI providers

The Context window control set to Auto, with a line beneath it reading Auto unavailable, using 32,000-token fallback.
The line under the control is the interesting part. It tells you whether the number came from your provider or from the fallback, which is a more honest arrangement than a built-in table of model sizes that goes stale.

Set a Custom size when you have a reason: a local model on modest hardware, or long threads you want compacted earlier. Managing conversations explains how compaction uses it.

Keep reading