Save the prompts that work and run them on any AI model

Put together, the email reply prompt looks like this:

Reply to the customer email below.

Goal: {{goal}}
Tone: plain and friendly, no exclamation marks
Length: under 120 words
Do not promise dates, refunds or features that are not in the notes.
If the email asks something the notes do not answer, say we will check and reply.

Notes:
{{notes}}

Email:
{{email}}

In AIHub, anything written in double curly braces becomes a field. When you run the prompt, the app asks for the goal, the notes and the email, fills them in and puts the finished prompt in the message box, so nobody has to find and replace text inside a long prompt.

A starter library of eight prompts

If you are starting from nothing, these eight cover most of what a small team hands to a model in a normal week. Write each one in your own words, with your own limits, rather than copying a generic version.

Name prompts by the job, not by the model

Name each prompt with a verb and a job, such as "Reply to customer email" or "Summarise in five points", so it can be found by typing what you want to do. Group them by the kind of work, not by which model you first wrote them for.

That second point matters more than it looks. A prompt that has been tuned against one model for months picks up small workarounds for that model's habits, and it quietly stops working as well anywhere else. Keep saved prompts plain: say what you want, give the limits, show the format. Then test each important prompt on two models at once, which is the quickest way to tell a good prompt from a prompt that only suits one model. The full method is in how to compare AI models side by side.

How the prompt library works in AIHub

1. Save it once. Create a prompt from the Prompts panel, give it a name and a category, and write the text with {{placeholders}} for the parts that change. When you use it, a small form asks for each placeholder and fills it in.
2. Find it by name. Search the panel by name or category, or type / in the message box to pick a saved prompt without leaving the chat.
3. Run it on any model. The prompt goes to whichever model is selected: GPT-6, Claude, Gemini, Llama, Mistral, Perplexity, DeepSeek or Grok. Your library does not change when the model does.
4. Compare two answers. Add a second model with the plus button and the same prompt goes to both, with each reply showing what it cost in credits.
5. Improve it without losing it. Every save is kept as a version. Switch the prompt editor to advanced mode to see the earlier versions and set any of them back as the live one, so a change that makes answers worse can be rolled back.

The library is included on every plan. Running a saved prompt costs exactly what typing it would, charged in credits at the selected model's price. On the Pro plan at $19 a month you get 10,000 credits, and a short message on Claude Sonnet 5 costs 7 of them. If you are weighing this against paying for several assistants, the numbers are laid out in ChatGPT Plus alternative: one subscription for every AI model.

Keep the library alive

Once a quarter, sort the library by what you actually used. Delete anything untouched for three months, merge near-duplicates, and rewrite the two prompts that most often need editing after the answer comes back. When a lab ships a new model, run your five most-used prompts on it next to the model you use now before you switch; ChatGPT vs Claude vs Gemini covers which jobs each one tends to win.

A library that gets this treatment stays short and trusted. One that never gets pruned turns into a folder nobody opens, and the team drifts back to typing the same request by hand. For where saved prompts fit in a small team's day, see where AI actually lands in a small business.