Made for the way you work
Direct connections, local work
Your workspace is the center of the work. The agent runs there, and your own provider keys connect it to the model you choose.
In practice
A useful place to start
Ask for a change in a local project. The agent reads and edits files on that machine, runs its tools there, and sends the needed context to your chosen cloud model.
Your machine does the work
The agent loop, file edits, terminal commands, and local search run on the workspace host. For an SSH or WSL workspace, that host is the remote machine or Linux environment you chose.
Your keys, your provider
Own-key model requests go from your workspace machine to the provider, without a Pumpkin-hosted inference relay. Keys use the editor's protected secret storage rather than project files.
Choose connected features
Cloud models receive sent prompts and context. Optional Pumpkin credits use a hosted relay. Remote uses encrypted device connections with relay fallback; sync uses your chosen storage. Connected tools and reporting follow their own settings.
Get started
Connect your own provider account in Settings. Review the services you enable, and choose local search embeddings if you want search indexing to stay on the machine.
Good to know
Pumpkin still uses online services for accounts, licensing, downloads, and enabled features. Local execution does not mean that cloud inference is offline or that nothing ever leaves your computer.