Variables

[DRYN07]

First of all, 👏 👏 👏

I've been using HARPA.AI since its inception, and it's one of the best AI extensions on the market. About 1-2 months ago, I switched from Chromium-based browsers to Zen, which has been a huge step forward in productivity. I can't imagine going back until I find a better option on Chromium. My biggest issue is the lack of AI plugins; it's insane that there is no such tool for Firefox-based browsers.

My second major problem was that Firefox-based AI tools lack variables (such as page/selected text awareness) and a custom prompt library.

As I see it, your plugin includes most of these features, which warms my heart and makes me very excited.

In these early stages, there are only a few things I miss:

1. The ability to work with selected text. In ChatGPTBox, there's a small popup menu, or you can use the context menu. Could you please consider a solution for this? I use prompts to fix grammar, translate, and fix code on selected text daily.

2. HARPA.AI has a really cool function called switches, where you can instantly indicate to the chat what you want:
   -p page awareness
   -w web awareness
   etc.

So if I want to chat with the page, I just write:
-p write SEO tags
-p summarize in 300 words
-p translate the first block

This is an immensely useful feature!

3. Single chat mode
   It would be nice to have an option to use the same chat across all browser tabs. I work on 10+ projects with about 200 tabs daily. When using the plugin, I'd like to see my prompts/answers on every tab where I open the popup. If I need something different, I can delete my chat (a reset button would be very helpful).

Overall, the effort you put into creating something functional makes me very happy, and the fact that you've made it available on Firefox is amazing.

If you have time (and aren't already aware of it), try out HARPA; there are lots of good ideas you could implement.

https://harpa.ai/guides/getting-started

[ctxinf]

@DRYN07 Thanks for using the extension and providing detailed feedback!

Regarding your ideas and needs:

1. I am currently working on it. The current design is: providing a context menu item for clicking to add the selected text to the chat box, after that, the user can manually enter additional prompt words and send the chat.

2. Currently, after summary step, every subsequent chat will send with the previous conversation as context messages to the API (same as what ChatGPT's web version and other clients do). This is equivalent to always using -p in HARPA.AI. Or do you mean manually control whether to include webpage content in the chat?

   (Chatting directly without first summarizing the page is expected to send messages with page context. But there is an bug now, causing direct chat will be a plain chat without page context. I will fix it soon.)

3. This idea sounds very interesting and innovative. The main challenge lies in context handling:

   1. complex webpage content is already quite long. With the current 128K context size (some of which must be reserved for output), accommodating multiple webpages is very hard.
   2. Online services can use RAG to reduce context size, but for webpage-summary, RAG or AGENT process was abandoned at the beginning of design. This is because they require vector storage and extensive interaction, which is nearly impossible for a purely serverless browser extension.

   I can think of 3 approaches :

   1. Each chat only includes the webpage content of the currently focused page along with the conversation history, while content from non-focused pages is not included in the context. This might match your need to work across different pages, but from a product design perspective, this usage pattern is not very intuitive.
   2. Implementing a mechanism similar to the -p command, allowing users to manually specify which webpage content to include. However, this would transform the extension from a simple summarization tool into a more feature-rich chat client. I may give it a try if there is certain ideas.
   3. An entirely different browser extension or tool that provides the MCP Model Context Protocol (github.com) service of accessing webpage content. You would then operate entirely within a dedicated MCP client.

[ctxinf]

@DRYN07 Hi, a context menu item for clicking to add the selected text to the chat has been implemented in 0.2.1!