[Question]: What I wanted to say/ask

[Snify89]

Describe your problem

Hi there,

this is for all the contributors, community members, etc. out there.
I hope it's alright, to compose a little longer Feature Request / Buglist.
First of all, I would like to thank you for all your hard work.
You are appreciated and hope/wish you all the best for your private lifes as well!
We know you are busy and have many things to do/answer/read/etc.
This is really hard work.
I love that Ragflow brings many nationalities, languages and characters together
because it works with so many languages which was a surprise to me.
As a German it is difficult to get good German language output from LLMs.
Suprisngly, not the west but the east (Asia?), offers better German output than llama, etc.
I came across glm4 which is fantastic for German output.
Anyway, I came across Ragflow and was astounded, by how easy it was to deploy and set this up.
But I was even more astounded about the backend mechanisms and I haven't even looked at Graph, tbh.
In short: This framework has strong potential and is exactly what I (and others) wanted/needed and it's easy to use.

So thank you and keep it up!

The project has a few hiccups here and there but it surely works fantastic and is also well documented.

There might be some duplicates in my list and I apologize, that I haven't looked for already other existing open requests/issues.

My feature requests:

• OpenAI compatible server API
  Many would instantly embedd Ragflow in their systems which would spread across businesses
  and gains more recognition. It would be also nice to pack everything into a "virtual" simple LLM.
  E.g. I would like to easily embeed a Ragflow Agent or Assistant (without the conditions like upload/options, etc.)
  and just use it as a simple model which has the power of a configured Ragflow in the back.

• OCRing with visual models
  Some visual models are doing fantastic in OCRing images and extracting content in tabular data style etc.
  This can help developers, to experiment easier with visual models and also use the generated content for searching.

• Agent (deploying)
  I had a brief look into the Graph and AI agent framework. It's awesome.
  It would be awesome to get like shareable links to the agents.
  The conditions for input(s) make it great for workflow distribution to contribute to the datasets.
  So others can login with credentials and pick a dataset, upload a file or files, etc.

• More language support
  Ragflow works/supports suprisingly well with German documents and many other languages and charts!
  When I use question/answer additionally, it sometimes creates "english" questions.
  Also, the OCR results misses "Sonderzeichen/Umlaute" like "ä", "ü".
  Sometimes you can edit prompts like in RAPTOR and stuff, which is good.
  But prompts for question/answer is missing or OCR language settings (e.g. if tesseract/tika is used)
  So I think there is room, to make this of worldwide/global interest.

• Add function calling, etc.
  I am not sure about the agents and GRAPH but it would be awesome to sort of function calling.
  There could be actions like HTTP Requests or even shell commands.
  Synchronous and asynchronous, etc.

• Add another upload API to file management (not just datasets)

Edit:

• Add Conversation/Memory functionality
  It would be nice to somehow use conversations workflows to store memories, etc.

These are my main features, I'd like to see in the future. So thanks in advance for considering :)

I have a few bugs/suggestions as well:

• Dark Mode is not working/applying

• Docker and documentation
  I have looked for ragflow and automatically got to the DEV documentation/images.
  This is why I might had some issues. I used the v0.14.1 image but in settings it still shows v0.14.0
  I also thought, that the dev image is always populated with the latest commit changes but somehow it isnt?!
  Maybe I got stuff got mixed up but I am not sure which image to use and how to update.
  So the docker documentations can be improved, I think.

• I am also trying to setup infinity but I had no luck. (won't apply chunks to infinity somehow)
  In the docs, it is mentioned that its not supported in Linux_X64?!

• Chunk edit window
  Something weird is going on with the chunk edit window not retrieving the chunks.
  I can't tell for sure whats going on but this might also be an issue with version migration stuff.

Otherwise guys, this is the real deal you made here and I am thankful for your
hard work and dedication to bring the world closer (by multi language support)!

Thank you for reading and considering any idea/suggestion.

[yingfeng]

Thank you very much for your attention and suggestions, support from the community is our greatest motivation to move forward!

Regarding to your proposals:

1. OpenAI compatible server API
   That's a very good offer. We'll consider it.

2. OCRing with visual models

We'll be adding more similar models, some trained by us and others from the community.

3. Agent

Agent is still in the process of improvement, it will not only have a shared connection, but also an Agent Store, so that different people can contribute Agents, each Agent is like an App, and RAGFlow is like an App Store.

4. More language support

Globalization is the goal of RAGFlow. We are indeed understaffed, but we will continue to make improvements to make RAGFlow truly global!

5. Function calling

There is an Invoke in the Agent, somewhat similar to function calling, fill in the URL, then you can call other services in the Agent. Any further requests for this feature are welcome!

6. API

There is still a lot of room for improvement in the API, and the file management API will be provided.

7. Memory management

Memory management is very important and we intend to use ES/Infinity to manage memory and provide real-time search (not just vectors, but full text, etc., which are indexed in real-time). Providing powerful multi-agents based on RAGFlow is one of the goals for next year.

[Snify89]

Thank you very much. This sounds promisable :)
I have another thing:
Parsing control is a little too limited. I'd like to autparse upon dataset entry or so and an option, how many maximum parsing threads (actually documents) can be processed at a time. The bulk option is ok, but I'd like to do this automatically with a configurable maximum on how many documents are processed.

Thank you.

Let me know, if you want me to do seperate issues for the suggestions/bugs.

Edit:
What about "metadata"? Is keyword(s) the replacement/alternative? Is it sufficient enough to store something like "Document type: Invoice" or "Invoice" ... ?

[yingfeng]

Such control could be provided through API in future.