Remember when I first started using Claude Projects? It felt like I'd finally unlocked the true promise of AI. I had a finely-tuned assistant that could co-write with me week after week, slashing my daily newsletter writing time from 2 hours to just 45 minutes. Easy street, right?

Well, not quite. After a few Playbooks, I got the dreaded "this conversation is running long" warning. Yikes. It all came crashing down. Suddenly, I was faced with a daunting question: Did I need to go through the whole process of priming, uploading, and narrowing every single time? What a pain that would be!

But don't worry - I've found solutions, and today, I'm going to share them with you. We're going to talk about how to overcome the context window limitation and make your AI assistant work portable for repeat use.

Let’s get started:

The Context Window Conundrum

First, let's address the elephant in the room: AI assistants, including those created with Claude Projects, have a limited context window. There's only so much information an AI can hold in its "memory" during a single conversation. After an amount of time or interaction, your carefully crafted assistant becomes less effective as it hits this limit.

This limitation can be frustrating, especially when you've spent time crafting the perfect assistant. But understanding this constraint is the first step to working around it.

Creating Portable AI Instructions

The simplest solution might seem to be starting a new chat and uploading the same information to retrain the AI. But that's slow, and the whole point of using AI is to save time, right?

Instead, let's capture all the work we've already done in a portable format. We'll ask Claude to create an output summary of everything it's learned. Here's a prompt to help with this:

You are an AI assistant that has been trained and refined for a specific task during this Project.

The below instructions refer to the conversation we have been having, not to your underlying model.

I need you to create a portable summary of your training and learning so that you can be quickly recreated in a new chat. Please provide the following:
1. A concise summary of your primary role and responsibilities.
2. The key instructions and guidelines you've been given for performing your task.
3. A summary of the main feedback and corrections you've received during the narrowing process.
4. Any specific techniques or approaches you've learned to be particularly effective for this task.
5. A list of common mistakes or pitfalls you've learned to avoid.
6. Instructions for how a new instance of you should request and incorporate the necessary knowledge base files. Ask for the same files as before.

Please format this information in a clear, organised manner that can be easily copied and pasted into a new chat to quickly recreate your capabilities.

Vital here is the second line. This tells Claude explicitly that we want to export it’s learnings from our work and not the underlying Claude model.

Without this Claude gets very protective and will start telling you it can’t divulge such information! It may still do so - if it does adjust the prompt so that it knows you are referring only to the current Project and not trying to extract it’s corporate secrets!

Another limitation is that Claude doesn’t have knowledge of what files you uploaded. So, upload all the same files again in the new chat instance.

Redeploying Your AI Assistant

Once you have this portable instruction set, here's how to use it:

1. Create a new Claude chat.

2. Paste in the portable instructions you generated.

3. Claude will likely ask for the data files it needs to get back up to speed. Upload these as requested.

4. Do a test run - you'll find the AI is already performing at a high level because it’ll have taken into account the feedback from the previous instance.

5. You may need to do some minor refinement, but it will be much less than starting from scratch.

This method allows you to quickly "reboot" your AI assistant when you hit the context window limit, without losing all the valuable training and narrowing you've done.

This works but still requires spinning up a new Assistant every once in a while. It’s inefficient.

Building Apps Around Your AI Assistant

For a more permanent solution, consider building an app around your AI assistant. This approach allows you to hook into Claude's API while maintaining your own local storage for instructions and data.

There are two main routes you can take:

1. No-code solution: Use a platform like Bubble.io to build your app without needing to write code.

2. Off-the-shelf solution: Use a service like Launch Lemonade, which provides ready-made AI app infrastructure.

The basic idea here is we will take our portable instructions from the past prompt and then ask how to build that Assistant using our platform of choice. Our portable instructions become a “scope document” in effect.

Here's a prompt to help you get started with creating instructions for a Bubble.io app:

You are an AI expert with experience in creating applications using Bubble.io. Please provide a high-level set of instructions for creating an AI assistant app in Bubble.io that uses the portable instructions above to quickly recreate the AI assistant's capabilities. Include the following in your instructions:

1. Initial setup steps in Bubble.io
2. How to create a user interface for inputting and displaying the portable instructions
3. Steps for connecting to Claude's API
4. How to implement storage for the portable instructions, including:
   a. Primary role and responsibilities
   b. Key instructions and guidelines
   c. Knowledge base file information
   d. Feedback and corrections summary
   e. Effective techniques and approaches
   f. Common mistakes and pitfalls
5. Workflow for:
   a. Inputting the portable instructions
   b. Initializing the AI with these instructions
   c. Sending user input to the AI and displaying responses

Please provide these instructions in a clear, step-by-step format suitable for someone with no experience with Bubble or AI API integration.

The goal is to create an app that allows users to quickly recreate their finely-tuned AI assistant using the portable instructions

Use this below the portable instructions to convert into a project task list. Adjust as required based on what tool you want to use and your comfort with development and APIs.