# Local versus hosted models

Ethen supports both **local** and **hosted** model execution. Which one you use
depends on your privacy requirements, hardware, latency tolerance, and model
quality needs.

## Comparison table

| Factor | Hosted (Gateway) | Local |
|---|---|---|
| Model selection | 100+ provider-hosted models via the Gateway catalog. | Open-weight models only (Llama, Mistral, Phi-3, Qwen, etc.). |
| Quality ceiling | Access to frontier models (Claude, GPT-4o, Gemini). | Smaller, less capable than frontier models. |
| Latency | Variable — depends on provider and plan. | Deterministic — depends on local hardware. |
| Privacy | Request data leaves your machine. | All inference stays on-device. |
| Setup | API key only. | Requires Ollama or LM Studio installation. |
| Cost | Per-token billing from providers. | Free (compute cost only). |
| Offline | Requires internet. | Fully offline-capable. |

## When to use hosted (Gateway)

Use the Gateway when you:

- Need frontier model quality (Claude, GPT-4o, Gemini).
- Want multi-provider fallback for reliability.
- Prefer zero local setup.
- Are building an API-facing service.

## When to use local

Use local runtimes when you:

- Must keep all data on-device (sensitive or regulated workloads).
- Need deterministic latency in a controlled environment.
- Work offline frequently.
- Are prototyping without cloud provider costs.

## Managing both

Both local and hosted models appear in the same Console interface. You can
switch between them freely per request. The model picker shows:

- **Gateway models** with a cloud icon.
- **Local models** from your running Ollama or LM Studio instance.

Local models you have loaded are available in the Console chat and the Code
workspace alongside hosted models.

## Getting started with local models

1. Install [Ollama](https://ollama.ai) or [LM Studio](https://lmstudio.ai).
2. Pull a model (e.g. `ollama pull llama3.2`).
3. The model appears automatically in the Console model picker.

See the Gateway [Quickstart](/docs/gateway/quickstart) for hosted setup.