Skip to content
Build Perch
Browse guides

How to run a bigger model without a local GPU

Short answer: Connect Ollama Cloud: your ollama.com subscription’s hosted models, run on Ollama’s own servers and proxied through Build Perch. It’s a flat-rate way to use large open-weight models (gpt-oss, DeepSeek, Qwen, and others) without a GPU capable of holding them locally.

Steps

  1. Create an API key at ollama.com/settings/keys.
  2. In Build Perch, open the Integrations window and select Ollama Cloud under LLM Providers.
  3. Paste the key into the Ollama Cloud row and save. Once saved, the key is never displayed again: it’s stored securely on your machine and used only to authenticate your requests to ollama.com.
  4. Open the model picker and switch the provider dropdown to Ollama (Cloud).
  5. Pick a model (Level 1, the lightest, is listed first and expanded by default), then click Connect.

Notes & gotchas

  • Billed by your ollama.com plan (Free/Pro/Max), not per token; the model picker shows “Subscription” instead of a $/token rate.
  • Models are grouped by usage level 1–4 as a rough size proxy: Level 1 is lightest and most likely to run on a free plan; Level 4 is heaviest and almost certainly needs a paid one.
  • It’s a separate provider and credential from local Ollama: connecting one doesn’t touch the other, and you can run both side by side in the same workspace.
  • No media generation. Like local Ollama, Ollama Cloud serves chat models only; image, speech, music, and video generation are OpenRouter-only.
  • If you already have a capable local GPU, running the same class of model through local Ollama instead costs nothing per message. Ollama Cloud is for when your hardware can’t hold the model you want, not a general replacement for it.