Alex takes us through the Aisle prompt library - a shared, multi-model prompt library that lets your team build, run, and version AI prompts across every major model.
A shared prompt library is where your team stores the prompts that are doing actual work day to day - sales email formatters, brand-language converters, bill-of-order templaters, translators, internal Q&A helpers, whatever your team has built. Aisle gives you one place to write those prompts, share them with the right people, and run them across ChatGPT, Claude, Gemini, Grok, and the rest without rewriting them for each model.
In the video I walk through building a prompt with variables, picking the model and toggling per-prompt settings (web search, citations, structured outputs, connectors, thinking), running it from the chat launcher, sharing it across the org or exposing it on an API endpoint, the revision log, optional usage logging, and the generate-with-AI shortcut for skipping the blank page.
The prompts your team writes are doing real work - drafting emails, structuring data, taking judgment calls that used to live in a person's head. Treating them as strings copy-pasted between Google Docs is fine until it isn't: someone leaves, the model changes, a prompt drifts, nobody can find the canonical version. Aisle gives prompts version history, access control, model portability, and logging - the basics any other valuable artifact in your business already gets.
Video transcript
Hi, Alex here, the founder of Aisle. And today I wanted to take you on a quick tour through our prompt library feature, which lets you create and share prompts with your team. Whether those run on chat GBT Claude Gemini Grock it doesn't matter. Those prompts might help you format sales emails, convert brand language, create something, you know, a bill order to a template, translate text, it doesn't matter. It's actually one of the origin cases of IO was the ability to build prompts and share them with your team in an environment they could run them in.
And recently been showing you some of our power user features like deterministic AI workflows with Python and tasks or these agentic and sub aent research processes with projects. It's good to reflect on this origin case again. So to get that started, we're going to click on prompts here. For now, we're going to do it manually. You can always press this generate with AI button.
Describe what you want. We're going to come back to that at the end. But so we understand the feature mechanically, we're going to come in and create a new prompt here. And we'll give this a name. Let's call it demo prompt.
We could always add a description if we want to make it easier for users to understand what it does and click a custom icon. So we put in our prompt information. This is what you might be copy pasting into chat today or copy pasting out of Google Docs or using some other tool for. We're going to it's really simple and say you know tell me a three-word joke where I can't even type on a demo joke about AI, right? And sometimes what we might do inside of a prompt is pass in variables, right?
things that we're going to change every time like the sales email, the data. What I do here is say about this topic, and I'm going to put in curly brackets here, the topic. You can have as many variables as you would like. Now, as we're building a prompt, you'll notice it has access to the date. And we can pick which model this is going to run on.
And we can change this later. We might run it on set at one point and change it to GPT later for speed, performance, who knows? Any of the attributes that are available on that model, you can toggle on and off here. Maybe we give this access to web search. Maybe we give it access to citations, turn on structured outputs, connectors, it's up to us.
Toggle the thinking settings, whatever we would like. Once this prompt has been created, we have the ability to run this inside of chat. So, I'm going to come back here and say, let's just pick something here. Why don't we talk about something like um, you know, RAM, memory, prices, just to get a basic demo running. And you'll see this is an easy way that we can share this prompt with our users and have them type in the details that matter and it will continue to run.
I can actually come back to this chat launcher at any time now and find this prompt here in my library. So we can make it easy for all of the users to find it. Have this wonderful environment to start building and sharing prompts with our team. Now to come back to this prompt for a moment and explain some of the more power features here. You'll notice that there is this sharing and deployment tab where you can control who in the organization has access to the prompt just yourself everybody or share it with specific people or you can even expose this prompt on the API endpoint if you'd like to call it from code.
Anytime we make a change to that prompt, it's going to appear in this revision log if that's changing the text, the attached model, the settings. This is how we build prompts that survive 6 12 18 months in production, right? We need to start maintaining them. And it is an optional feature you turn on and off if you would like logging on your prompt to be able to see things like token count, what went in, what went out. This is all conditional stuff you can turn on and off.
But that gives us this nice structured entity for your prompt. In my mind, prompts are the modern IP, right? This is what you're building. It's replacing parts of your business. You can platform them, start to share them with your team and maintain them, which is what we must do.
Hopefully, this is helpful. I said we'd do an AI one at the end. So, let's click back to the prompts index. And most of the time we don't handwrite these prompts anymore. What we would do is come in here and press generate with AI and I might say write a prompt that reads an AI prompt and provides feedback based on best practices.
Now most of the time we would want to provide those best practices or do a whole bunch of extra information. But you'll see this will take us to a chat section and it's going to start generating that prompt for us. Once this prompt has been generated, we can actually sit here inside of this chat window and provide feedback to it. If there are things we don't like, we'd like it to change and we can start to see a diff of that. But we'll just sit here and this prompt has come through.
So, we could ask for changes. Otherwise, once this has been made, you have two choices. You can press create prompt and it's going to take you here to save it and put whatever details you would like. Or you can press playground and start running different versions of this in different models, which is something that we will cover in a later video. If, for example, you would like to run this on Sonic 46 and 48 and see how they both perform, that's something you can do here.
Anyway, thank you for watching.