> ## 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.

# Iwant quickstart

> Install Iwant, launch `gpt-oss-20b` on GCP, query its endpoint, and tear down the GPU.

You need the `gcloud` command-line tool and a GCP project with billing enabled, L4 GPU quota, and available capacity. Installing from source requires Python 3.9–3.14.

## Install and authenticate

Download and run the installer:

```bash theme={null}
curl -fsSL https://github.com/narevai/iwant/releases/latest/download/install.sh -o /tmp/iwant-install.sh
bash /tmp/iwant-install.sh

gcloud config set project YOUR_PROJECT_ID
gcloud auth login
gcloud auth application-default login
iwant auth
```

The installer uses `~/.iwant` and places the command in `~/.local/bin`. Add `~/.local/bin` to your `PATH` if your shell cannot find `iwant`. To disable anonymous installer telemetry, run the installer with `IWANT_NO_TELEMETRY=1`.

## Preview and launch

Inspect the plan without provisioning resources, then launch:

```bash theme={null}
iwant up gpt-oss-20b --infra gcp --dry-run --yes
iwant up gpt-oss-20b --infra gcp --yes
```

The recipe requests one L4 GPU. Save the **Server** URL and **API key** printed by the launch. Provisioning, model downloads, and loading can take time. Iwant checks that the models API responds before reporting readiness.

## Send a request

Set the URL and key from your launch:

```bash theme={null}
export OPENAI_BASE_URL="http://<IP>:8000/v1"
export OPENAI_API_KEY="<generated-key>"

curl --fail-with-body "$OPENAI_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-oss-20b","messages":[{"role":"user","content":"Write a haiku about GPUs."}],"max_tokens":256}'
```

Use the same URL, key, and model ID in your OpenAI-compatible application.

## Remove your deployment

Cloud resources keep billing until you remove them:

```bash theme={null}
iwant down
iwant list
```

Select the cluster to remove and confirm it no longer appears. For scripts, run `iwant down EXACT_CLUSTER_NAME` using the name from `iwant list`.

> [!WARNING]
>
> Ctrl+C detaches from a launch; it does not stop the remote server. The default 30-minute idle timer tracks cluster jobs, not API requests. A running model server keeps the cluster active. Run `iwant down` when you finish, including after an interrupted launch.

See [launch options](https://www.narev.ai/docs/iwant/usage.md) for region selection, spot capacity, and recipe versions. This guide follows the upstream [quickstart](https://github.com/narevai/iwant#try-gpt-oss-20b).


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