Open Feedback: How are you using Mark?

[relston]

I would love to get some feedback from anyone who has use mark. All comments, questions, concerns, anecdotes, stories, and rants are welcome!

Some starter questions:

• Are you using mark? Why or why not?
  • What have you used it for?
  • How has it been helpful?
• Did you have any trouble with mark?
  • What frustrated you?
• Is there a task your trying do with an LLM currently?
  • Does mark help with this task?
• What other LLM tools are you looking at?
  • Which have you tried? What did you think of them?
• What is the single most important feature missing from mark?

Answer any question you want! It's not a survey, just looking for your thoughts.

[philScholl]

Hi Ryan,

Kudos for mark - I have checked out and tried it, it's a nifty tool and I really like the workflow / UX, as I try to organize all my work in text files and this is a great fit (I don't have too much use for image generation right now). I keep all my LLM conversations in different files and can index / grep them to find them again if needed.

However, I do have very specific requirements (set of different LLMs, both local and company-internal with no OpenAI-compatible API, mixing conversational with tool and rag use cases), so I cannot use mark without major changes right now. I have done some adaptations locally, but I had to rip out and change too much so only the file structure and the name remains... and it's too bad code to contribute back right now :)

Still, mark is a great tool, thanks for your work!

[relston]

Thanks @philScholl for responding and for the feedback. I really appreciate it. I agree that the workflow/UX is one of the chief benefits of this tool. I always feel that I have the LLM constantly at my fingertips.

I'd like to know more about your specific use cases and workflows. What specifically did you change to make it work for you? I definitely figured that being limited to OpenAI or OpenAI-compatible endpoints is going to be a deal breaker for a lot of people. So, I've been thinking about ways to expand this elegantly. I was planning on looking into Ollama support as a potential next step.

https://github.com/ollama/ollama

As it has an API that is fairly similar to OpenAI, I imagine there are plenty of client libraries available for it. Would something like this fit your use cases.

Thanks again for your response!

[johnblommers]

Mark has been a handy way to generate images.

Moving forward I'm going to see how good Mark is at extracting or generating tables from web sites, kind of doing AI web scraping without using those "free with premium options" subscriptionware tools like Kadoa, Bardeen, Browse AI, WebScraping.AI, ScrapeStorm and Anypicker.

[relston]

Thanks for the reply @johnblommers
Let me know how it works for your use case. My scrapper logic is pretty basic but I have found it gets the job done for most things I throw at it.

The obvious limitation is it's not going to hand any auth or user session, so it wont read any content that requires a login 🤷

But I don't think most other web scrapping services will do that either though. Maybe you can correct me on that?

[johnblommers]

But I don't think most other web scrapping services will do that either though. Maybe you can correct me on that?

No web scrapers I've encountered have supported login/authentication, and frankly it's never been a use case of mine.

[relston]

Right on. Actually, now that I think about it, browser plugin based scrapper will obviously be able to use your session. But yeah, those usually make it pretty awkward to do scrapping at a large scale.

Anyway, there is certainly room for improvement with the scraper so let me know if you run into any issues @johnblommers
Cheers!