Run local models with Ollama
Ollama runs open AI models directly on your Windows PC. It is the free, fully offline, fully private way to power SideSense. Here is how to set it up.
What Ollama gives you
Ollama is a small program that downloads and runs open-weight models on your own hardware. Paired with SideSense it means your drafts are generated on-device, with no API key and no usage cost. Nothing about the session touches the network.
Install Ollama on Windows
Download the Windows installer, OllamaSetup.exe, from ollama.com and run it. It installs without administrator rights and finishes in about a minute. Ollama then runs quietly in the background and adds an icon to your system tray, near the clock.
Pull a model
Open a terminal (Windows Terminal or PowerShell) and pull a model. The first pull downloads the weights, which can be a few gigabytes depending on the model, and then it is cached for offline use. For example:
ollama run phi4 # small and fast, good on modest PCs ollama run gemma3 # balanced quality and size ollama run llama3.3 # stronger, needs more memory
Current open models worth trying include Phi-4, Gemma 3, Qwen 3, and Llama 3.3. Any model in the Ollama library works; these are just good starting points.
Select it in SideSense
Open Settings and go to AI models. Choose Ollama as your provider and pick the model you pulled. From then on SideSense drafts your answers with that local model, fully offline. You can switch models any time, or add a hosted provider key if you would rather use one.
Picking a model for your machine
Larger models are smarter but need more memory and run best with a GPU. Smaller models are quicker and lighter. If drafts feel slow, drop to a smaller model; if your PC has the headroom, a larger one gives better answers. It is worth trying two or three to find the balance you like.
Questions, answered.
Is Ollama free?
Yes. Ollama is free and open-source, and the models it runs are open-weight, so there is no per-request API cost. You only need disk space for the model and a machine that can run it.
Which model should I pick?
It depends on your PC. A smaller model like Phi-4 or a small Gemma 3 runs comfortably on modest machines, while a larger Llama 3.3 or Qwen 3 is stronger but needs more memory and ideally a GPU. Start small, then size up if your machine keeps pace.
Does it really work offline?
Yes. Once a model is downloaded, Ollama runs entirely on your machine with no internet connection. On this lane nothing about your session leaves your PC at all.
Free, private, and fully on your machine.
Install Ollama once and run SideSense with no API key and no cloud.