# iwant

> iwant deploys open models on GCP by finding available GPUs, provisioning a machine, and setting up an OpenAI-compatible endpoint. It is for developers who want to run models in their own cloud account without managing infrastructure.

- Type: AI tool
- Page: https://completeaitraining.com/ai-tools/iwant/
- Published: 2026-10-09
- Updated: 2026-10-10
- Author: Jeroen Erne
- Jobs: IT and development
- Topics: Developer Tools, Open Source
- Reading time: 2 min

### About iwant

iwant is a command-line tool that automates the deployment of open AI models on Google Cloud Platform (GCP). It locates available GPU capacity across regions, provisions the infrastructure, downloads model weights, and configures a vLLM serving endpoint. Users receive an OpenAI-compatible API endpoint, an API key, and SSH access - all running within their own cloud account.

### Review

iwant tackles a specific friction point: turning a model experiment into a working API often means manually checking GPU availability across zones, handling provisioning scripts, and configuring inference servers. The tool compresses those steps into a few commands. It's a focused utility that doesn't try to be a full MLOps platform.

#### Key Features

- **Automated GPU hunting:** iwant scans GCP regions to find available GPU capacity, eliminating manual zone-by-zone checks.
- **One-command model deployment:** The `iwant up` command provisions a machine, downloads weights, and sets up vLLM with a single instruction.
- **Model recipes:** Pre-configured recipes abstract hardware requirements and serving settings for specific models like DeepSeek, Kimi, Minimax, MiMo, and Step.
- **Cluster management:** `iwant list` shows running deployments and their connection details, while `iwant down` tears down clusters.
- **Dry-run inspection:** The `--dry-run` flag lets users review the planned infrastructure before committing resources.

#### Pricing and Value

iwant itself is free and open source. Users pay only their own GCP infrastructure costs - compute, GPU, and networking charges billed directly by Google Cloud. The tool works with existing GCP credits. No pricing tiers or paid plans are mentioned in the available information.

#### Pros

- Runs entirely in the user's GCP project, keeping data and infrastructure under their control.
- Reduces multi-step GPU provisioning to a single command.
- Exposes an OpenAI-compatible endpoint, so existing client code works without modification.
- Model recipes handle hardware selection, removing guesswork about GPU type and count.
- Source code is open, allowing inspection and modification.

#### Cons

- GCP is the only supported cloud provider at launch; AWS and Azure users cannot use the tool.
- Users need pre-approved GPU quota and billing enabled - the tool doesn't help obtain or increase quotas.
- The tool is not well suited for teams running production workloads that require autoscaling, persistent monitoring, or multi-node orchestration beyond what a single vLLM instance provides.

iwant fits developers and researchers who want to test open models quickly without writing Terraform scripts or clicking through cloud consoles. It's most practical for short-lived experiments, demos, and evaluation workflows where spinning up and tearing down GPU instances on demand matters more than long-running service management. Teams with multi-cloud needs or strict production requirements will likely need additional tooling alongside it.

---

This is the Markdown form of https://completeaitraining.com/ai-tools/iwant/. The whole site is mapped at https://completeaitraining.com/llms.txt.