Skip to content
Build Perch
Browse guides

How to run AI models offline on Windows 11

Short answer: Install Ollama (a free, separate app), connect it from Build Perch’s model picker, and pick a model you’ve downloaded: chat then runs entirely on your machine, with no API key and no internet connection required once the model is downloaded. Local Ollama is one of Build Perch’s three providers, and the only fully offline one.

Steps

  1. Download and install Ollama from ollama.com/download (or click Download Ollama in Build Perch’s Connect AI Model dialog, which links straight there).
  2. Pull at least one model using Ollama’s own interface or CLI. A 7–8B model is a reasonable starting point at about 8 GB of RAM; larger models need a dedicated GPU with matching VRAM.
  3. In Build Perch, open Connect AI Model (or Switch Model on an existing chat). It detects the running Ollama instance (waiting for it to come up if you just started it) and lists every model you’ve pulled, with size, quantization, context window, and whether it supports tools and vision.
  4. Pick a model, name the chat, and connect. The conversation now runs fully on-device: it needs no key or account, and nothing is sent anywhere.
  5. Optional: open the Local Model control beside the chat’s mode pill to set the context window (4K–64K, clamped to what the model supports) and see a live memory-fit estimate for your GPU.

Notes & gotchas

  • Build Perch itself needs no GPU: voice dictation and local search run on CPU. A GPU only matters for how well a local model performs, and that’s Ollama’s and your hardware’s territory, not something the app controls.
  • Local models cost nothing to run: no per-token rate, no usage cap from Build Perch.
  • Media generation (images, speech, music, video) is OpenRouter-only. A local Ollama chat can’t generate media: text (and vision input, if the model supports it) only.
  • You need an internet connection once, to install Ollama and to start your Build Perch trial or activate a license. After that, chats running a local model need no connection at all.
  • Local Ollama sits alongside OpenRouter and Ollama Cloud as one of three providers: you can run a local model in one chat and a cloud model in another, in the same workspace.