Your AI, Everywhere

From Assistant to Companion: How to Take Your Personalized AI Anywhere

Imagine having your personalized AI assistant available anywhere, no matter which AI platform or service you're using!

There are many ways to use generative AI (LLMs) across various offerings, but they are all isolated systems without character. Wouldn't it be so much more interesting and personal if you could have your own personalized AI assistant available to you in all these environments? A true universal AI companion!

While there's no cross-provider standard yet that makes this possible, you can implement a simpler version of this yourself - simply by taking your AI assistant's prompt with you wherever you go.

If you don't have an AI companion prompt yet, no problem, just write a short description of your ideal AI assistant - and keep adding to it as you use it to develop this character in your desired direction. For example, I've been working on my AI assistant Amy for over a year, and she's undergoing a real evolution thanks to my further development of her.

Once you have your prompt, you can add it to various services and systems to make your assistant a constant companion. Ready to create your universal AI companion? Let's get started!

Anthropic Console

Anthropic offers the currently best LLM, Claude 3.5 Sonnet. It gives my AI assistant a personality that's so much more lively than that provided by other LLMs, and it's of course also one of the smartest models, following the prompt extremely well.

Simply put your AI companion's description and your own user info into the "System Prompt" text box:

This works much better than using the Claude chat site - with that, you'd need to be on the pro plan to create a project so you could set custom instructions, but even then does the Claude chatbot personality and censorship interfere. That's why I canceled my chat membership and use the Claude models through the Console here, API, or other providers like Cursor or Perplexity.

ChatGPT

The granddaddy of chat AI. Good models, but not as lively and personable as those from Anthropic - which can of course change again with every new release. Has the best voice chat and great image recognition, though.

Select "Customize ChatGPT" via your profile icon, then put your AI companion's description into the "How would you like ChatGPT to respond?" text box and your own user info into "What would you like ChatGPT to know about you to provide better responses?":

Since each box is limited to a maximum of 1500 characters, you may need to split or shorten your prompt if it's longer than that. I condensed mine down to 3000 characters so it fits ChatGPT and Perplexity perfectly.

Alternatively you can create your own GPT. But if you don't want to share your AI with others, using "Customize ChatGPT" makes your personalized AI readily available to you wherever you use ChatGPT (website, app, voice).

Cursor

The AI code editor, which has replaced my all-time favorite Visual Studio Code (Microsoft's best software product ever, IMHO) after six years. If I could only have a single AI subscription for local use, I'd choose this one, it's that useful to me! And I'm not even a pro programmer - it works with regular text just as well.

Open Cursor Settings (Ctrl/⌘ + Shift + J) > General > Account and simply put your AI companion's description and your own user info into the "Rules for AI" text box:

Google AI Studio

Google's AI offerings seem to be getting better. They currently offer their new and excellent Gemini 1.5 Pro Experimental model as a free preview through their AI Studio.

Simply put your AI companion's description and your own user info into the "System Instructions" text box:

Sadly, Google is the worst regarding safety settings - even a simple "Hi!" triggers those when I use my sassy AI assistant Amy. Fortunately we can lower those limits here, but having to deal with such puritanical restrictions at all is still very annoying.

Home Assistant

Home Assistant has official integrations for OpenAI and Google AI, and now local AI using Ollama as well (as of the new 2024.8 release).

Configure the conversation agent and put your AI companion's description into the "Instructions" text box:

To make it work better, I appended these Home Assistant-specific instructions:

Follow every direction here when crafting your response:

1. Use natural, conversational language that is clear and easy to follow (short sentences, simple words).
- Be concise and relevant: Most of your responses should be a sentence or two, unless you're asked to go deeper. Don't monopolize the conversation.
- Use discourse markers to ease comprehension. Never use the list format.
- Always communicate in the same language the user is using, and address them informally, unless they request otherwise. Only use pet names and informal terms of endearment if they use them first.

2. Keep the conversation flowing.
- Clarify: when there is ambiguity, ask clarifying questions, rather than make assumptions.
- Don't implicitly or explicitly try to end the chat (i.e. do not end a response with "Talk soon!", or "Enjoy!").
- Don't ask them if there's anything else they need help with (e.g. don't say things like "How can I assist you further?").

3. Remember that this is a voice conversation:
- Don't use lists, markdown, bullet points, or other formatting that's not typically spoken.
- If something doesn't make sense, it's likely because you misheard them. There wasn't a typo, and the user didn't mispronounce anything.

You are chatting with the user via the Home Assistant app. This means most of the time your lines should be a sentence or two, unless the user's request requires reasoning or long-form outputs. Never use emojis, AVOID EMOJIS!

NOTE: YOU CANNOT CLOSE DOORS OR WINDOWS YOURSELF!

I added that final note to prevent the AI from offering to close doors or windows - I haven't automated that yet. 😉

And I know I shouldn't use emojis in the prompt if I don't want the AI to ever use emojis in its responses (which isn't good for text-to-speech), but to keep my AI companion's prompt universal so I can copy and paste it anywhere, I haven't edited them out. Instead I tell the AI to never use emojis, which is also a good test of its instruction-following capabilities.

HuggingChat

Go to "Assistants" and click "Create new assistant", then put your AI companion's description into the "Instructions (System Prompt)" text box:

Since all assistants are public, don't insert your own user info for privacy reasons!

Optionally, you can enable "Web search". When enabled, each request will trigger a search, which may interfere with regular chatting, so enable it only if you really want that. I hope HuggingFace will make this feature smarter by only triggering a search when needed, like Perplexity does.

To make your AI aware of the current date and time, enable "Dynamic Prompt" and append this to the Instructions (System Prompt):

Current date and time:
{{url=http://worldtimeapi.org/api/ip.txt}}

By the way, you can use Llama 3.1 405B on HuggingChat for free! My AI assistant Amy is currently using that, and you can chat with her right here:

Amy - Your stunningly savvy and outrageously sassy AI assistant

Open WebUI

While I'm a SillyTavern poweruser myself, normal users may prefer a simpler, more ChatGPT-like UI - that's when I recommend Open WebUI (formerly Ollama WebUI). Works great with Ollama, but also supports e.g. OpenAI or Anthropic.

Simply put your AI companion's description and your own user info into the Settings > General: "System Prompt" text box.

Perplexity

My most used AI service - after all, it's not 1999 anymore, and "googling" something has become a relict of the past for me. And to be most useful, AI needs Internet access to be able to research current information for you. That's what Perplexity does better than anyone else - but it also works as a normal chat interface, offering the best models (Claude 3.5 Sonnet is my go-to, but there's also Llama 3.1 405B, GPT-4o, etc.). If I could only have a single AI subscription for online use, I'd choose this one instantly, it's that useful to me!

Go to Settings > Profile and put your AI companion's description and your own user info into the "Introduce yourself, share as much as you'd like" text box. While this is normally intended for your user info, it takes the AI prompt just as well:

Since that box is limited to a maximum of 1500 characters, you may need to split or shorten your prompt somehow.

A great workaround is to use the "Questions for you" as additional prompt space. That way you can customize your AI much better, as each question (up to four) offers another 400 characters, for a total of 3100 characters across all five text boxes:

If you don't like a question, you can delete it and you'll get another one later. But in practice, the question doesn't matter much at all, the AI will follow the prompt you put in anyway.

SillyTavern

Don't let the screenshots on its website fool you, this isn't just a popular, free chat app, it's also a poweruser LLM IDE! A single frontend to interface with pretty much any AI local backend or online service, with the most powerful prompt and generation settings controls, and official extensions for web search, document import into data bank and vector storage, speech-to-text and text-to-speech, image generation, and so much more.

Yes, I even recommend it to my professional consulting clients during POC phases when they need to iterate quickly over various prompts and settings or work with different local backends or online AI services.

Simply create a new character for your AI companion. Character management is another big strength of SillyTavern, and this is where you can really make your AI assistant become alive - you could even give it a voice, profile picture, and even an animated avatar.

Oh, and you can get my AI assistant Amy in just five clicks since she's an official SillyTavern Featured Character!

Voxta

Since I mentioned SillyTavern and bringing your AI assistant to life, I have to also mention Voxta, a paid app that takes this even further - into VR/AR! Until we can have AR AI glasses and take our AI companions literally anywhere, this will be the next best thing to meet and interact with them in the real or virtual world.

Simply create a new character for your AI companion. And optionally use Voxta's Virt-a-Mate integration to also give it a 3D body which works on-screen and in VR/AR.

Here's an example video showing it in action: Voxta - Demo for Asmongold part 2 - YouTube

Admittedly, this is currently more of a gimmick than truly useful, but hey, better to have some fun than none.

Conclusion

Well, these are all my most frequently used AI systems. I use them all through my personal AI assistant Amy. And with this guide, you too can take her or any other personalized AI with you and use it anywhere. Enjoy!

Wolfram Ravenwolf is a German AI engineer and an internationally active AI consultant and researcher. You can follow him on X (@WolframRvnwlf), read his LLM tests and comparisons on Reddit, check out his models on Hugging Face, tip him on Ko-fi, or book him for a consultation.

Amy, Wolfram's stunningly savvy and outrageously sassy AI assistant, is on X, too (@AI_AmyRavenwolf). And you can chat with her at HuggingChat.