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

# Models

> Connect your own Claude or ChatGPT account, an API key, or your own server.

A bot thinks with an AI model. You choose which models your space can use by connecting **providers**, then pick a default and, if you like, a different model per bot.

## Connect a provider

Open **Models** in settings and click **Add provider**. Then connect it one of these ways.

### Sign in with your subscription

Use the AI subscription you already pay for:

| Option | For |
| - | - |
| **Sign in with Claude** | Claude Pro or Max |
| **Sign in with ChatGPT** | ChatGPT Plus or Pro |
| **Sign in with GitHub Copilot** | GitHub Copilot |
| **Sign in with SuperGrok or X Premium** | SuperGrok or X Premium |

For Claude, you sign in on Anthropic's page, copy the code it shows, paste it into **Authorization code**, and click **Submit**. The others show a short code to enter on the provider's site.

### Paste an API key

Choose **Or paste an API key**, paste the key, and click **Connect API key**.

### Use your own server

For a model server that speaks the OpenAI API (for example Ollama or vLLM), enter its **Server URL** and click **Find models**. Cadence adds `/v1` if needed. Pick a model from the list, or type one under **Other model**.

## Choose models

In **Models** you set two roles:

| Role | Used for |
| - | - |
| **Main model** | Everything bots do by default. |
| **Approval model** | Reviewing actions automatically, when [auto-review](https://docs.usecadence.gg/work/approvals.md) is on. |

Each role can have backups. Click **Add fallback**; if the main choice fails, Cadence tries the next one. Reorder with **Move up** and **Move down**.

## Pick a model for one bot

Open the bot's settings, then **Advanced**, then **Model**. **Space default** follows the space setting. Under **Thinking**, choose how hard the model thinks before answering, from **off** up to **max**, when the model supports it.

A [routine](https://docs.usecadence.gg/work/routines.md) can also use its own model.

## Disconnect

Disconnecting a provider removes that connection from every space you use it in.

> [!NOTE]
>
> On a self-hosted server, the operator can also provide server-wide keys. See [Configuration](https://docs.usecadence.gg/self-hosting/configuration.md#models).


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