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

# AI models

> The Rerun models on your OpenRouter key, your own API keys, and your Claude or ChatGPT plan.

Every coworker thinks with one model at a time. On a self-hosted Rerun you pay the AI provider directly: there is no credit, no meter and no plan in between.

## The Rerun models

A new coworker starts on **Rerun**, with three tiers to pick from:

| Tier | For |
| - | - |
| **Fast** | Quick, simple jobs |
| **Balanced** | Everyday work. The default |
| **Smart** | The hardest tasks |

On your install, these run on your `OPENROUTER_API_KEY`. The engine asks the app, the app relays the call to OpenRouter with your key, and nothing is counted. All three tiers are open to every workspace.

The app only relays for engines it knows, so your key never serves anybody else. The Foreman and its helpers run on the same key.

> [!TIP]
>
> Set a credit limit on the key in your OpenRouter dashboard. It is the simplest way to cap what your coworkers can spend.

Without `OPENROUTER_API_KEY`, a coworker on a Rerun model answers with an error asking you to set the key or connect another provider. Connect one below and pick it, and the coworker works again.

The OpenRouter model behind each tier is set in `RERUN_UPSTREAM_MODELS`, in `apps/ui/lib/engine/constants.ts`. Change it and rebuild to run a tier on another model.

## Your own API keys

Open the workspace menu, the pill in the top left corner, and click **AI providers**. Add a key for any of these:

| Provider | Gives you |
| - | - |
| **Anthropic** | Claude models |
| **OpenAI** | GPT models |
| **Google** | Gemini models |
| **OpenRouter** | Every model OpenRouter carries. Also accepts a custom base URL and custom model ids, to point coworkers at your own router |

Provider credentials belong to the workspace. They are stored on the workspace's server, in its data-dir, and never in Supabase. Every coworker of the workspace can use them.

## Your Claude or ChatGPT plan

In the same **AI providers** dialog, connect a plan you already pay for:

* **Claude**, with a Pro or Max plan
* **ChatGPT**, with a Plus or Pro plan

Sign in once and your coworkers run on the plan. The engine image ships the official Claude Code and Codex command-line tools that make this work, so there is nothing to install on the server.

## Choose the model per coworker

Click the model name above a coworker's message box. The picker lists the Rerun tiers, your plans and your keys, with a power rating and the context window of each model.

Picking a model in the chat changes it for that conversation. Click **Set as default** to change it for the coworker. See [Configure a coworker](https://docs.rerun.build/agents/configure.md#model-and-provider) for the details.

## Embeddings

Semantic search of the services catalog and of the coworkers' memories needs embeddings. They run on `RERUN_EMBED_API_KEY`, an OpenRouter key, through the app. Without it, both searches fall back to keyword search, which works but matches less.

You can use the same OpenRouter key for both variables, or a separate one to see the two costs apart.


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