Quickstart

Five commands take a container from your laptop to two or more GPU providers behind one URL.

Install
Terminal window
```
uv tool install multihull
```
Initialise a spec

hull init detects a Dockerfile, vLLM or TGI setup and writes multihull.yaml.
Terminal window
```
hull init
```
Check credentials and GPU availability
Terminal window
```
$ hull doctor
gke-prod    OK
modal-main  OK
runpod-eu   OK  (L4 in EU-RO-1)
```
Credentials come from each provider’s native location (kubeconfig, ~/.modal.toml, RUNPOD_API_KEY). Nothing goes in the file.
Plan

Renders every provider’s native payload to .multihull/plan/ and diffs it against recorded state.
Terminal window
```
hull plan
```

Deploy and watch

$ hull deploy
$ hull status
gke-prod    Ready  2/2  L4    https://gke.int/llama
modal-main  Ready  1/1  A10G  https://acme--multihull-llama-8b.modal.run
runpod-eu   Ready  1/1  L4    https://api.runpod.ai/v2/abc/

All targets apply concurrently. One failing target never blocks the others.

Run the router: see Router overview or the Helm chart in charts/multihull.
Prove failover: hull failover test -p gke-prod drains the primary for 60 s and reports traffic shift and error count.
Example spec: examples/llama-8b/multihull.yaml in the repository.

Quickstart

Next