Made for the way you work
A model on your machine
Keep a model close to your files. Pumpkin supports a local model for chats and small tasks that help the editor run.
In practice
A useful place to start
Let the local model help name chats and describe saved data. Save your cloud model choice for tasks that need more capability.
Local inference
The local model runs on your computer. Once its files are installed, that model's inference does not need a cloud model endpoint. Available speed depends on your processor, memory, and supported GPU.
Useful background help
Pumpkin can use the local helper for small requests, such as naming chats, describing data, and choosing a model for a task. You can choose a hosted substitute if your machine is too slow.
Visible setup
Settings shows install progress and lets you install, cancel, or remove the model. Local prompt profiles keep the starting context and tool set suited to a smaller model.
Get started
Open Settings, then Models, and find the local model. Install it on a machine with enough memory and disk space, or choose a helper substitute.
Good to know
A local model has different capabilities from frontier cloud models. Downloads, account checks, web tools, and any cloud substitute still use the network. Local inference alone does not make every feature offline.