# Switch to another AI

Canonical: https://godspeedmissioncontrol.com/guide/switch-ai

Godspeed thinks with the AI account set in Hermes, the program that runs your assistant. Change it there, and the chat and your routines both follow. The notebook's chat shows the new model within five minutes, after you reload the page.

## With an API key, in Settings

1. In your notebook, open **Settings**. **Your AI** is near the top.
2. Choose the service, fill in **Model** and **Key**, and check that **Use it for the daily routines and the chat too** is ticked.
3. Click **Test and connect**. It sets Hermes as the steps below do, then checks that your routines reach the new AI.

## With a ChatGPT plan, in Settings

1. In **Settings**, under **Your AI service**, choose **ChatGPT (sign in with your ChatGPT plan)**.
2. Click **Sign in with ChatGPT**, open the page it shows, type the code there and sign in.
3. It says "Connected. Your ChatGPT plan" with the model it found.

## On Windows

In your notebook, click **Settings**, then **Open the Windows assistant**. The Hermes app opens: change it in the app's settings, under **Model**. Or use PowerShell, which asks step by step:

1. Press the Windows key, type `PowerShell` and press Enter.
2. Paste this line and press Enter:

   ```powershell
   $s = "$env:LOCALAPPDATA\Godspeed Mission Control State"; if (-not (Test-Path "$s\installation.json")) { $s = "$env:LOCALAPPDATA\Godspeed Mission Control Full Alpha State" }
   $a = Get-Content "$s\assistant.json" -Raw | ConvertFrom-Json; $env:HERMES_HOME = $a.home; & $a.executable model
   ```

   The first line picks your state folder. A Windows install made before 2.0.0-beta.11 keeps the older name, which ends in "Full Alpha State", also after updating.

3. Carry on from step 2 of [On a Mac or Linux computer](#on-a-mac-or-linux-computer). Its lists are numbered: type a number and press Enter.

## On a Mac or Linux computer

1. Open a new terminal window and run:

   ```bash
   ~/.local/bin/hermes model
   ```

2. It shows the AI it uses now, then a list of AI services. Choose with the arrow keys and Enter.
   - **A ChatGPT plan:** choose **OpenAI**, then **ChatGPT or Codex Subscription**.
   - **An API key:** choose the company your key is from.
3. Sign in:
   - **ChatGPT:** open `https://auth.openai.com/codex/device` on any device, type the code it shows and sign in.
   - **API key:** paste your key when it asks.
4. Under "Select default model:", choose a model.
5. You see a line that starts "Default model set to:". Close the window.

## On a server

1. In your Telegram chat with Godspeed, send `/model --global`. (`/model` alone changes it for this chat only.)
2. Pick a model from the buttons.

Telegram exists only on a server installed with the one-line command, and its buttons offer the AI services already set up there. On a server without Docker, follow the Mac and Linux steps in its terminal.

## For one chat only

Under the notebook's message box, **Model** picks another model and **Reasoning effort** (with a ChatGPT plan) sets how hard it thinks, until you close the chat.

## Related

- [Connect a backup AI](/guide/backup-ai)
- [First setup: connect your AI](/guide/first-setup)
- [Chat commands](/guide/chat-commands)