> ## Documentation Index
> Fetch the complete documentation index at: https://docs.intellikon.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Prompts
> Create a prompt, try it in the playground, save new versions and go back to an earlier one.
**Prompts** lists every prompt in your workspace, with its product, type, status and current version. Each prompt keeps a full history: every time you save a change to the prompt, model or settings, Nucleus adds a new version and keeps the ones before it, so you can compare them or go back. Use the search box and the product, type and status filters above the list to find a prompt, then select it to open it.
## Before you start
Tenant Owners and Workspace Admins can create prompts, go back to an earlier version and turn a prompt off. Workspace Members can open, edit and try them. Every prompt is filed under a product and a type, which Tenant Owners and Workspace Admins set up in [Nucleus settings](/products/nucleus/settings).
## Create a prompt
Select **Prompts**, then **New prompt**.
Enter a **Prompt Name** your team will recognise.
Choose the **Product**.
Choose the **Type**.
Add a **Description** if you want one. It shows under the name in the list.
Under **Content**, write the instruction.
Put anything that changes from one use to the next in double braces, such as `{{ticket}}`.
Select **Create Prompt**.
To put the first version in **dev** or **staging** straight away, choose **Dev** or **Staging** under **Environment** before you create it. Otherwise leave it at **None**.
The prompt now appears in **Prompts** as Active. Next, [try it in the playground](#try-a-prompt-in-the-playground).
To insert a variable without typing the braces, select **Variables**.
## Try a prompt in the playground
The playground sends the prompt to the AI model and shows you the answer, without changing the prompt.
Select **Prompts**, then the prompt.
Select **Run in playground**.
Enter a value for each variable the prompt uses.
Select **Run**.
The answer appears under **Output**, with the model used and how long it took. Each run is a real request to the model. Next, [save a new version](#save-a-new-version) if you want to change the prompt.
## Save a new version
The **Model** list shows the models your workspace allows for Nucleus. Tenant Owners and Workspace Admins manage them in Studio settings, under **AI Models**.
Select **Prompts**, then the prompt. The newest version opens in the editor.
Edit the text under **Prompt** if you want to change it.
Change the **Model** if you want.
Change the **Temperature** if you want.
Change the **Max tokens** if you want.
Enter a **Commit message**, such as "Asks for a reason with each priority".
Select **Save**.
The new version appears at the top of **Version history**, marked **Current**, with your message and your name. Next, [compare it with the version before](#compare-two-versions), or [release it](/products/nucleus/releasing-prompts).
Select **Expand** above the prompt to edit it full screen. Press Escape to close it.
## Compare two versions
On the prompt, select **Compare**.
Choose a version in the first list.
Choose a version in the second list. Nucleus highlights what changed in the prompt, the model, the settings and the commit message.
Select **Split** to see the versions side by side, or **Unified** to see them in one column.
Select **Exit**.
You are back in the editor. Next, [go back to an earlier version](#go-back-to-an-earlier-version) if the older one was better.
While comparing, press the up and down arrow keys, or J and K, to jump between changes.
## Go back to an earlier version
Select any version in **Version history** to read it. Older versions are read-only. Tenant Owners and Workspace Admins can bring one back.
In **Version history**, find the version you want back.
Select **Revert** on that version.
Confirm that you want to revert.
Nucleus adds a new version with the older content and makes it current, so nothing in the history is lost. Next, [release it](/products/nucleus/releasing-prompts).
## Turn a prompt off
Tenant Owners and Workspace Admins can turn a prompt off and on again.
Select **Prompts**, then the prompt.
Select **Deactivate**.
Confirm that you want to deactivate it.
The prompt now shows as Inactive. It stays in your list with all its versions.
To turn it back on:
Select **Prompts**, then the prompt.
Select **Activate**.
Next, [check the health of your active prompts](/products/nucleus/monitoring#check-prompt-health).
## Troubleshooting
Change the prompt, the model or a setting, and enter a **Commit message**. Nucleus only saves a version that differs from the one before.
A new prompt has no evaluation yet, so it has not passed the quality check for **staging**. Write why it should start there, or choose **Dev** instead.
Another active prompt in your workspace has that name. Choose a different **Prompt Name**.
The name belongs to one of Studio's built-in prompts. Choose a different **Prompt Name**.
Ask a Tenant Owner or Workspace Admin to add a type for that product in [Nucleus settings](/products/nucleus/settings#add-a-type).
Your workspace has no models allowed for Nucleus yet. Ask a Tenant Owner or Workspace Admin to allow one in Studio settings, under **AI Models**.
Only Tenant Owners and Workspace Admins can create prompts. Ask one of them to create it, then you can edit it.