> ## Documentation Index
> Fetch the complete documentation index at: https://www.narev.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Launch options and recipes

> Choose Iwant models, regions, spot capacity, server keys, and deployment lifecycle options.

## Commands

| Command | Purpose |
| - | - |
| `iwant up [MODEL]` | Launch a model, or open the interactive model picker. |
| `iwant list [CLUSTER]` | Inspect deployments and endpoints. |
| `iwant down [CLUSTER]` | Remove a deployment, or pick one interactively. |
| `iwant auth` | Check cloud credentials and show login guidance. |
| `iwant ssh [CLUSTER]` | Open SSH to a running cluster. |

Use exact cluster names from `iwant list` in scripts. Run `iwant up --help` for the complete option list.

## Launch configuration

```bash theme={null}
# Preview without renting a GPU.
iwant up gpt-oss-20b --infra gcp --dry-run --yes

# Choose a region.
iwant up gpt-oss-20b --infra gcp/us-central1 --yes

# Request interruptible spot capacity.
iwant up gpt-oss-20b --spot

# Authenticate a model download. Gated models also require granted access.
HF_TOKEN=hf_xxx iwant up gpt-oss-20b

# Hide streamed setup logs.
iwant up gpt-oss-20b --quiet
```

Use `--api-key` to supply a server key, or let Iwant generate one. GCP is the currently supported cloud. Quota does not guarantee capacity in your selected region.

## Models and recipe versions

Use `iwant up` to select an available model. The [upstream model table](https://github.com/narevai/iwant#models) lists GPU requirements and verification notes. Experimental configurations are not guaranteed to launch successfully in every environment.

`iwant up MODEL` selects the highest `v<N>.yaml` recipe in that model's recipe directory. Launch output records a label such as `step-3.7-flash@v2`; pass the model name, not that label, to the command-line tool. A recipe version does not freeze every upstream dependency.

For another model, discover its served identifier before sending a chat request:

```bash theme={null}
curl --fail-with-body "$OPENAI_BASE_URL/models" \
  -H "Authorization: Bearer $OPENAI_API_KEY"
```

## Lifecycle and costs

```bash theme={null}
iwant up gpt-oss-20b --idle-minutes 60
iwant up gpt-oss-20b --no-autostop
```

The default idle timer is 30 minutes. It tracks cluster job activity; a running server keeps the cluster active even without incoming requests. `--no-autostop` disables the timer. Always tear down resources explicitly when finished.

After an interrupted or failed launch, inspect `iwant list` and remove any unwanted cluster with `iwant down EXACT_CLUSTER_NAME`. Hourly costs vary by region and capacity type. Inspect the plan and cloud rates before provisioning.

These instructions follow the upstream [usage guide](https://github.com/narevai/iwant/blob/main/docs/usage.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.