"Personalized ChatGPT" is a good way to describe what the Alice app is at its core. At first glance, it looks like any other chat app, but the magic happens when you switch between assistants and activate their skills, called "Snippets". But let's start from the beginning.
When you send a simple message within the Alice app, it'll be sent to the large language model that's currently active (you can see its name in the bottom-left corner of the window).
Besides your message, the system prompt of the currently active assistant will be sent, affecting the model's behavior.
You can start a new chat by pressing Command + N (or Control + N on Windows) or by clicking the New Chat button.
All your conversations are stored locally on your device for 14 days. You can access them again through the History panel by pressing Command + Shift + H (or Control + Shift + H). If you decide to remove a conversation, it will be gone forever, and neither I, Greg, nor Adam will be able to restore it.
Near the main prompt input, you'll notice information about the currently active model. You can switch it by pressing Command + M and navigating with the arrows over the available models.
On the right, there's information about the currently active provider and the available token limit for the current conversation. If you go beyond that limit, your requests will be rejected because of the model's limitations.
It's recommended to keep threads short because each time you send a message, the entire conversation is sent to the language model, and you pay for all those tokens. Also, the model's performance gets worse when you provide it with too much information at once.
The Alice app lets you upload images (PNG, GIF, JPEG) and files (TXT, MD, CSV, and PDF) that are no bigger than 10MB and contain no more than 75,000 words. Be cautious, though, because if you want to work with bigger files but your active model has low input token limits, the request will fail.
Uploaded files and images are represented as below, so it's either an icon of the file type or preview of the image.
Files that are waiting in the queue will be added to the message you send. So, for example, the contents of a text file will be added to the end of your message, and the active assistant will be able to read it.
By default, Large Language Models aren't connected to the internet. Apps like ChatGPT, Claude, or Alice change this by performing additional steps between your message and the assistant's response. These steps are responsible for:
To activate Online Mode in Alice, you'll need to register for a free account on the firecrawl.dev and add your API key in the Settings -> General tab.
When your API key is set, you can activate Online Mode by pressing Command + O (or Control + O on Windows). Then, the globe icon will highlight, and your assistant will use the web search tool whenever you ask about websites or recent info.
Here, I have an answer related to the hardware mentioned on brain.overment.com. Besides the answer, there's also a list of references used to generate it.
If you want the option to generate images, you'll need an account on Replicate.com and a billing method set up.
WARNING: Make sure you've set a monthly spending limit.
Then you can generate API key and add it within the Settings -> General tab.
Similar to Online Mode, you need to activate Image Mode by pressing Command + I (or Control + I on Windows). Then, simply tell me what you'd like to see, and you'll get an image generated with the imagegen-3 model available here: https://replicate.com/google/imagen-3.
The Alice app lets you interact with different language models. How an assistant responds depends on two things: the model itself and the system prompt of the assistant you're using. You can change the system prompt in the assistant's settings.
Stay aware that many limitations and the general quality of the conversation depend on the model you're using and the active prompt. The more you understand how Large Language Models work, the better you'll be at working with the Alice app.