---
question: "Should I run ABI locally, with Docker, or on Kubernetes?"
---

# Choose your path

The same project runs in more than one way. Pick the run mode, the module kind, and where the model runs. The pages linked here are the procedures.

## Where it runs

| You want | Use |
|---|---|
| To develop on your machine, without Docker | [`abi dev up`](/user-guide/get-started/quickstart) |
| The full stack on one machine | [`abi deploy local`](/user-guide/deploy/local-docker) |
| One server with a domain and certificates | [`abi deploy dokploy`](/user-guide/deploy/dokploy) |
| Your own cluster, or Naas | [Production](/user-guide/deploy/production) |
| No outbound network | [Air-gapped](/user-guide/deploy/air-gapped) |

Install requirements are on [Install and prerequisites](/user-guide/get-started/install). A comparison of the deployment shapes is on [Deployment models](/user-guide/deploy/deployment-models).

## What you add to the project

| You are building | Create |
|---|---|
| An agent, an app, or a scheduled job in its own process | `abi new module <name>` (remote module) |
| Pipelines, workflows, integrations, or ontologies inside the engine | `abi new module <name> --in-engine` |

The choice, and what each kind can contain, is on [Creating a Module](/user-guide/build/modules/creating-a-module).

## Where the model runs

A new project uses a cloud model through OpenRouter. To run a model on the machine instead, use Ollama. Both are on [AI mode: cloud vs. local](/user-guide/get-started/quickstart#ai-mode-cloud-vs-local).

The canonical ids of those models are `services.model_registry.default_chat_model` and `default_embedding_model`. See [Choose the default model](/user-guide/build/models/choose-the-default-model).