Choose the model and capacity
Set the model, GPU size and session duration in chalupa.yml. Preview the estimated hourly rate before launching.
chalupa upPrivate compute. Your cloud account.
Give your model more room. Connect a remote GPU to the tools on your laptop over SSH. Run temporary Compose environments and CI jobs with the same CLI.
curl -fsSL https://chalupa.run/install.sh | bashInstalls bun, go-task, and the CLI for you. Then run chalupa setup to connect your cloud account, SSH key, and console account.
inference:
model: qwen3.8:27b
pull: true
contextSize: 65536
maxModelBytes: 25769803776Configuration excerpt · requires a session and local expiry scheduler
Serve open weights on your own host and connect your local tools through SSH.
chalupa launch opencodeYour infrastructure. Direct provider billing.
GPU inference · experimental
Choose an open-weight model served by Ollama. Chalupa provisions the GPU host and connects your agent through a local endpoint.
Connect your agentSet the model, GPU size and session duration in chalupa.yml. Preview the estimated hourly rate before launching.
chalupa upRequest a context window or negotiate one against the model ceiling and observed GPU placement. Weights, context and runtime all need memory.
With hosted reporting configured, inspect GPU memory and utilization alongside reported tokens, speed and turn outcomes. Then stop compute with chalupa down.
A short-lived server. A familiar workflow.
Chalupa handles the environment lifecycle through the CLI. Your editor, tools, and cloud account stay yours.
Start with your Compose file, a CI suite, or an inference configuration. Preview the resolved plan offline.
chalupa previewConfirm the launch, then open an SSH tunnel. Service ports bind to localhost on the remote host.
chalupa tunnelUse your environment. The optional hosted console keeps status, estimated costs, and reported results in view.
chalupa statusTear down compute explicitly. Declared persistent data lives in its own stack and continues to be billed separately.
chalupa downClear boundaries
Access, storage, and lifecycle each have a defined place. The console observes your fleet; launch and teardown stay in your CLI.
Explore the operating modelService ports stay on localhost. SSH is the entry point, scoped by your configured network allowlist.
Protected volumes live in a separate stack. Stopping compute preserves declared persistent storage.
Launch and teardown ask for confirmation. Inference expiry needs an awake Mac and unlocked credentials; it is not a hard spending cap.
Free CLI. Optional hosted visibility.
Keep provider billing in your account. Hosted plans add fleet history, CI evidence, and cost visibility. GPU time and model downloads are separate.
Compare plansThe complete local CLI and TUI for your own infrastructure.
Hosted operations and visibility for an individual developer.
More capacity, collaborators, and history for small teams.
Billing beta: usage is recorded; prepaid packs and overages are not enabled.
Start where you already work