Write with web search and working with enabled web search

[slu-010101]

Try this prompt: "Do web research and write a colorful, detailed summary of the space elevator, suitable for physicists."
next try: "for English teachers."
run "soffice --writer" from the command line so you can see what's going on in the background.
Also you can click on the web research chat button and paste in a web link and ask it to summarize the web page.

@KeithCu yes I tested it, it's absolutely amazing!

First I tried it with enabled "Web Research" but this generate only a result in the chatbox, right?
Without enabled "Web Research" the text was added into my document.

[KeithCu]

Glad you liked it!

Yes, "Web Research" is just to chat directly with the research agent. It doesn't know about your document or anything. I changed the text in the prompt to be clearer: "AI: I can do web research to answer any question, or summarize a web page, without seeing or changing your document. Let's chat."

I might change the checkbox to "Research Chat" instead so it makes it clearer. Would that be a better name, what do you suggest?

[KeithCu]

Okay I did full localization into a bunch of languages including German. Try out the new release if you want and let me know how it goes.

[KeithCu]

Hello @slu-010101! when you have a minute, do you want to try out the onboarding librarian assistant and memory feature? Also I made some other changes to support progressive disclosure of the full UNO API: https://github.com/KeithCu/writeragent/blob/master/docs/writer-specialized-toolsets.md

If it works for you i'll push this new build to the extensions website with the full language translation and the onboarding agent and the other work I've been doing behind the scenes. No rush, just try it out when you have time and see what you think about it ;-)

I didn't update the version number, but you'll see the notes in the release.

Thanks for all of your feedback!

[slu-010101]

do you want to try out the onboarding librarian assistant and memory feature?

Onboarding means "Evaluation Dashboard"?
The latest version doesn't open the "Evaluation Dashboard" nor "Settings":

[KeithCu]

Oops I fixed the settings dialog box and put up a new version. Sorry and thank you! The eval dashboard is a test feature, ignore it for now.

This is text I just put into the Readme: Librarian onboarding agent (memory-aware): For new users without a config/Memories/USER.md, the sidebar starts with a Librarian / Welcome sub-agent that chats with the user to learn preferences: name, favorite color, etc. When the Librarian (or main agent) decides something is worth remembering, it uses the upsert_memory tool. The memories are passed into requests automatically. When the Librarian / new user are done chatting and want to work, it switches control to the main document assistant for the current session and future ones.

It's basically a way to save your name, favorite color, etc. ;-)

[slu-010101]

Works, but it ask me now every time I start a new chat and the folder in ~/.config/libreoffice/4/user/config/memories is empty.
Shouldn't the settings be saved there?

[KeithCu]

Yes, it should be saving your settings to there. It must be failing creating a directory or writing to disk?! Can you please send me the logs?
You can just create a file like this to make it happy ;-)

USER.md
You can also put in your favorite color, which is why I added the feature in the first place ;-)

[KeithCu]

I may have figured it out. It's probably a pyyaml dependency issue which likely doesn't work on most systems. I've converted it to a json but storing it in the same md file ;-)

I also changed it so it gives more error messages in the chat when something unusual happens to make it easier for the user and me to figure out what's going on.

I'll just put a new release up there. Try it when you want. Thank you for testing! I have 540 tests but that still doesn't cover everything! This release has the Clear button translation fix too.

[KeithCu]

BTW, in the next build, I'm going to dump any errors into the chat sidebar to make it easy for the user and me to discover without digging through log files. This is still in a development / prototype stage so it's fine to show full error data whenever bad things happen, which shouldn't be too often, but it happens more often on other machines ;-/

[slu-010101]

I may have figured it out. It's probably a pyyaml dependency issue which likely doesn't work on most systems. I've converted it to a json but storing it in the same md file ;-)

I can confirm the file was created.
He asked me about my favorite color, I answered, he confirm but the color isn't in the USER.md.

Actually only my name is in the file:

{
  "name": "Samuel"
}

No errors so far.

Thank you for all the work on this add-on!

[KeithCu]

Thanks very much for your reports. I'd have to see the logs to find out what happened but I'm very glad it doesn't keep asking your name. Agents are supposed to be useful, not annoying. Not remembering your favorite color is not a big deal even though it's why I added the feature ;-)

It's an interesting, challenging project which is why I keep working.

[slu-010101]

I tried it again with an empty LibreOffice profile, it's only the user name in the USER.md, nothing else.
As far I understand you, the color should be also in this file?
There is nothing in the writeragent_debug.log about this.

It's not important to me, I just wanted to give some feedback.

[KeithCu]

Thank you for the feedback! BTW, it's just the USER.md file it uses to decide if it should spawn the librarian agent to welcome you. It's okay, it's probably the models aren't always smart enough yet to do the updates for color. I'll change the prompt to help it, hopefully the system becomes more useful over time, based on the memory feature. The name and color are the best initial examples.

[KeithCu]

BTW, I just fixed the issue with the librarian agent going into document mode too quickly ;-) I'll put it in the next release.

[KeithCu]

Okay this should be fixed and work better and offer tips and try to guess that it's time to switch over to document mode.