I would like to connenct a custom built chatbot to a logseq graph

I would like to connenct a custom built chatbot to a logseq graph.
The idea is ,

  • the user would copy paste a link.
  • The app would visit the link and copy in the contents
  • the ai behind the app would parse the content and create a set of hashtgs (to become links in the graph)
  • the app would then pass the content, the set of tags and the url to to a graph/page which would do its magic.
    something something the user would use logseq for query etc. or the app with more AI would work with logseq for new insight and analysis

: Questions:

  • is this feasible?

  • will it need an API resource that is not yet available,

  • could it be done with existing logseq resources.

  • Where should I start?

  • is there place where app concepts are discussed and evaluated for feasibility?
    Thanks in advance.

On first reading, I don’t see any obstacle in essence. The various parts already exist, now someone has to code their integration. Though I don’t fully understand your description:

What do you mean by “magic”?

  • What will the app be developed with?
  • Who will do the coding?

By magic i was referring to what LS does when a user creates blocks, tags/pages, the linkage and back linkage mechanism that LS creates and makes available to a user, etc that I would like made available to an external automation based on NLP.

IOW to allow external code or an Agent (say), to work with LS as if it were a user.

The app could be developed with any code approach unless specific frameworks have already been set up (I’m new to logseq and have not explored all the resources) but typically for early AI/Conversational demos would be python and related libraries and prompt automation.
.
I;m thinking I’ll start the demo coding, but if I make acquaintance with a motivated partner, would be happy to share the development so I can focus on fund raising, pitching and client sponsorship.

Feel free to ask as much as you want…
Best

From an LLM perspective, what you call LS’ magic isn’t particularly magical. In theory:

  • an LLM includes a much richer (although incomprehensible) graph than the one in LS
  • unless some human intelligence intervenes in the process, every step away from the original LLM degrades the model’s quality

Overall, it is not clear what value is added to LLM, if LS doesn’t contain relatively original content.

Step two, web capture, seems great but difficult. It could be implemented as a plugin and would be added value even if the rest of your system doesn’t get developed.

3 and 4 need more description. Step by step, what is each thing that will happen? If you can do that, then you can get some help estimate the feasibility.

Thank you for your response

With so much change happening in the AI space I frequently revisit the aspects and approaches

As for web scraping or api interfaces, i’m considering a hive of agents that are specialists in various different techniques that could compete and inform each other on successful ways to handle any particular website each using its own approach that it’s an expert in (beautiful soup, response, selenium, etc ) until they figure out what works then report back to the executive who will pass it on to the logseq handler.

:man_shrugging:

The scope and utility of that is so large I’d suggest it be its own service. There is probably an existing service that provides an API to extract content from web sites, though I’m sure it’s not free.