2026.May.05: Proposal: split the functionality of the plugin into a part aimed at software projects and another for general matters

[SINAPSA-IC]

I opine that the LM Studio Toolbox could be split into two plugins:

• one for working on software projects, which I assume was its initial intent, containing:
  -- Git operations, finding-replacing text in (code) files, opening system's Terminal and executing commands there, and other such "heavyweight" stuff

• and one for more general functionality, like:
  -- browsing the web (info retrieval and RAG), RAG on local files of whatever nature, listing the contents of a directory, find the most recent file of type .xyz, get system info, copy-move-delete-rename files, and other "lightweight" functionality - that is, not related to coding, but separate from it.

Ofc, both could be joined in a... comprehensive plugin. At the present though, I, for one, would like to use a plugin aimed at frequent operations performed by all/common users, and if need be, install the other one which would be aimed at programmers and more advanced users.

The value of the functionality brought about by this plugin is so great, and it so advances the scope of LLMs in LM Studio, that it could very well profit from the hip room available - which IMO, is splitting the functionality and renaming the functions (now, the list of functions is unordered alphabetically, and their categories are not signaled in that list, so a specific function is usually hard to find.)

Also IMO, when dealing with AI there's the habit of building software with the advanced users of /AI in mind, instead of catering to the common users who account for about 70% of the user base (7 in 10 users of /AI are common users that only use email, MS Office and ChatGPT for homework and pancake recipes, and the other 3 have had some contact with coding.)

[Beledarian]

@SINAPSA-IC Do you have any naming suggestions? For the alternative one? And branch a common plugin of this one and keep this as software dev? Or the other way around? And how would you plan on maintaining/ keeping them aligned, to not have to submit the same pr twice/ to keep them aligned regarding the most important features? I could also see adding add-ons, if we could implement a system for that. By keeping this as is/ removing the gh tools or so and adding for a example a dev core + GitHub core plugin etc. but this would maybe lead to maintenance overhead I personally couldn't deal with alone.

And on the other hand there are a lot of great individual plugins for web search etc out there. It would be cool to unify them into an ecosystem maybe, but regarding this project here we shouldn't strip it of too many features as it would loose its intended Toolbox - eg. 1 click install -> have the full agentic/ llm toolsuite at your disposal.

[Beledarian]

Does anyone have any Input or Wants to be heared regarding this idea?
@LifetimeVip what do you think?

[SINAPSA-IC]

You said, "It would be cool to unify them into an ecosystem maybe"
I think it's not room for maybe anymore, at this stage of LLMs' extended functionality employed at large scale for various procedures, but it's imperative to profit from your work.
This plugin that you've put together is VERY useful (I daresay it's the MOST useful out there, with so many tools being aggregated like a Swiss-knife of LM Studio plugins - which is something the users want-or would like to use, anyway) in its current form; yet, an expansion of sorts (splitting it in two - one for daily/ordinary/basic functions and a more comprehensive form) would double-down on its impact.
Keep in mind that there's an arms race in this domain, and you still have the higher ground ref LM Studio plugins.

And now, back to the front.

Organizing the functions:

1. in the code file:

• its readability would be improved by separating them via several blank lines, or by several lines of blank comments

• place start and end comments at the start and the end of each category of functions, for instance
  //here begins the functions category: FILES
  ...
  ...
  //here ends the functions category: FILES
  //
  //
  //
  //
  //here begins the functions category: WEB OPS
  ...
  ...
  //here ends the functions category: WEB OPS

• arrange/order the categories alphabetically: CLIPBOARD would come before DIRECTORIES/PATHS, which would precede the category FILES

• in each category, arrange/order the functions alphabetically: file_copy would come before file_read, which would come before file_save

2. Naming (to ^that end)

• each function would be named top-down, from its object/domain ("clipboard", "file", "rag", "web" etc.) down to the action itself:
  read_document -> document_read_pdf_docx (not realistic an adequate name for additional file types, though)
  send_notification -> system_notification_send
  analyze_project -> software_project_analyze
  change_directory -> directory_change_chdir
  save_memory -> conversation_save_to_file (something from the conversation, I guess)
  save_file -> file_save
  replace_text_in_file -> file_replace_text_inside
  list_directory -> directory_list_dir
  read_file -> file_read
  execute_command -> system_command_run_shell
  make_directory -> directory_make_mkdir
  delete_path -> path_delete_file_or_directory
  delete_files_by_pattern -> files_by_pattern_delete
  run_in_terminal -> system_command_run_terminal
  web_search -> web_search (no change)
  move_file -> file_move
  copy_file -> file_copy
  fetch_web_content -> web_content_fetch
  rag_web_content -> web_content_rag
  get_system_info -> system_info_get
  find_files -> files_find
  fuzzy_find_local_files -> files_local_find_fuzzy
  get_file_metadata -> file_metadata_get (subject to expansion to multiple files)
  read_clipboard -> clipboard_read
  write_clipboard -> clipboard_write
  open_file -> file_or_URL_open
  preview_html -> html_preview
  browser_session_open -> browser_session_open (no change)
  browser_session_control -> browser_session_control (no change)
  browser_session_close -> browser_session_close (no change)
  browser_open_page -> browser_open_page (no change)
  run_test_command -> test_command_run
  wikipedia_search -> wikipedia_search (no change)
  rag_local_files -> rag_on_files_local
  consult_secondary_agent -> secondary_agent_consult
  insert_at_line -> file_insert_text_at_linenumber
  append_file -> file_append_content_at_end
  read_file_range -> file_lines_range_read
  search_in_file -> file_content_search_pattern
  delete_lines_in_file -> file_lines_range_delete

the Git/Hub functions -> similarly

->
(very hard to keep track of them all in ^that unordered order)

//browser
browser_session_close
browser_session_control
browser_session_open
browser_open_page

//clipboard
clipboard_read
clipboard_write

//directory (folder)
directory_change_chdir
directory_list_dir
directory_make_mkdir
path_delete_file_or_directory

//files
document_read_pdf_docx
file_copy
file_move
file_or_URL_open
file_read
file_save
files_by_pattern_delete
files_find

//file contents
file_append_content_at_end
file_content_search_pattern
file_insert_text_at_linenumber
file_lines_range_delete
file_lines_range_read
file_replace_text_inside
file_content_search_pattern

//RAG
rag_on_files_local
rag_on_web_content

//secondary agent
secondary_agent_consult

//software projects
analyze_project -> software_project_analyze

//system
system_info_get
system_command_run_shell
system_command_run_terminal
system_notification_send

//web
web_content_fetch
web_search
wikipedia_search

//miscellaneous
conversation_save_to_file
html_preview
open_file_or_URL
test_command_run

As for maintaining the functions after renaming them EVERYWHERE (in all files where) they appear, that's a non issue I think, because once given other names it will be the same situation as now; after all, what's in a name? (/quote) :)

And you will be doing that renaming - if you so want, ofc - or not, because you are the sole author of this set of functions; we the users will download the form of the plugin that you offer us.

[SINAPSA-IC]

Also, the functions activated ("Allow...") by default (or when the plugin is first loaded?) could be different in the advanced/complete version (like "Allow Git*.*" which might be a must there) than those in the general-purpose version (like "Enable Wikipedia", altthough Wikipedia may be useful in any situation). Not some world-ending stuff, but just saying...

[Beledarian]

@SINAPSA-IC First thank you for going into so much depth with you suggestions. Maybe we could somehow group tools by category and sort each category alphabetically? But the names should destroy the immediate understandability of the tools by name only.

system_command_run_shell
system_command_run_terminal

For example seems a little bit much. Especially as we would group tools by functionality extensions? Or would you just do 2 different toolsuites?

Second:
I can truly see the value of what we could do here but sadly im fairly time constrained due to the work I have to do in general. So while I'd love to be able to maintain multiple projects I dont think that I'd be able to sustain that workload. Motivationally for a open source plugin. And more important in general. We could most definitely enhance this tool suite and if you want I could make you a maintainer and wed maybe need a few more people engaging with this so that becomes worth it. I could imagine putting the newly added gh tools into a separate extension plugin and renaming/ sorting the tools. I also would add tools and functionality where I personally need them or someone requests it. But all of this would have to be gradual. And I certainly wouldn't feel comfortable aiming to maintain multiple plugins as I doubt I will stay motivated doing that on my weekends 😅

But in general, I would love to maintain this toolbox and make it relevant for casual users. Maybe a good first step would be separating the advanced and normal tools in the setting, putting the more casual tools at the top while putting the specialised at the bottom. In the toggle config, aswell as the explicit tool list.

What do you think?

[SINAPSA-IC]

_But in general, I would love to maintain this toolbox and make it relevant for casual users. Maybe a good first step would be separating the advanced and normal tools in the setting, putting the more casual tools at the top while putting the specialised at the bottom. In the toggle config, aswell as the explicit tool list.

What do you think?_

@Beledarian That's a good! idea, rearranging the tools, and especially! good is the option to select the set of tools in the activation menu of the plugin; yet all this involves a bit of extra work. You do as you wish, as it's a good idea, and I second this motion :)

As for the motivation and more people, I understand all too well, so I second this motion too. Anyway, I'll try and do these 2 things:

• write a Medium post about your plugin + LM Studio
• put some links leading to the plugin's pages here on GitHub, on my website (not a guaranteed large/decent audience, but still the possibility would exist that someone would stumble on it out there and awaken the dragon, as they say)
• I've already supplemented a tutorial on LM Studio that I've been writing and Augmenting since some 3 months ago, with the method to use your plugin, but that tutorial is for a Very small audience (since among the younger generation around here, the smartphone full of Insta. Filtered selfies and TikTok videos is more important than what's going on around them, like, in real life, so that guide of mine is specifically targeted to a handful of Interested youngsters in their 20s)
• I'm at a loss what more could I do to encourage the use of this plugin by the population at large. Well, I could ask an LLM... nah...

Be well! and bring us good news!

[Beledarian]

@SINAPSA-IC You mind sharing your tutorial and/ or post in a discussion here once its done? I'd be interested in your work

Thank you for being so engaged :)
Do you think this plugin is relevant for the population at large? I mean I personally I do enjoy using and experimenting with local llms in my free time. But for most (non-technical) people it seems like they don't even realize that this is something you can do

I'll look into it. Maybe I can try to use an agent to experiment with this. This seems like a refactor they should easily handle.

[LifetimeVip]

您好， 后来我又下载过最新版本的lm studio，运行了您的插件，但是由于我使用的pc是16g vram，而且lm studio的作者根本不来合并我更新的语言包，导致我看不懂英文，实际体验很差，现在我都降级用9b模型了，只用lmstudio帮针对截图做翻译使用。虽然之前我能够运行gemma4 26B A4b Q4km这样的模型，大概每秒50TOKEN，但是我认为指挥lmstudio这样的软件，根本不现实。这个studio在几个月之前应该是个不错的选择，但是现在我的主力软件都放在opencode和openclaw上面去了。我认为这才是未来。 如果LM studio不变成open code 这样的项目，没有什么价值。当然opencode和openclaw也应该加入支持本地模型进行驱动，因为很多人喜欢私密和隐私性。 我认为未来的路只有一条，就是输入语音或者文字就能驱动软体自我完成任务，期间会自我思考和修正。 如果您联系不到lm studio的作者，那么帮他做插件，没什么意义。不如尽早寻找新的项目。 欢迎一切交流。 祝您愉快 ---------------------------------- Hello, Later on, I downloaded the latest version of LM Studio and ran your plugin. However, since my PC only has 16GB of VRAM, combined with the fact that the LM Studio developers simply refuse to merge the language pack updates I provided (leaving me unable to understand the English interface), the actual experience is very poor. Currently, I have downgraded to using 9B models and use LM Studio solely for translating screenshots. Although I was previously able to run models like Gemma4 26B-A4b-Q4_KM at approximately 50 tokens per second, I believe it is simply unrealistic to try to command software like LM Studio in this way. This studio platform might have been a good choice a few months ago, but now my primary tools are OpenCode and OpenClaw. I believe this approach is the future. If LM Studio does not evolve into an open-code project like these, it holds no value. Of course, OpenCode and OpenClaw should also integrate support for local model driving, as many users prioritize privacy. In my view, there is only one path forward: inputting voice or text to drive software to complete tasks autonomously, involving self-reasoning and self-correction during the process. If you cannot reach the LM Studio authors, then developing plugins for them serves no purpose. It is better to seek out new projects sooner rather than later. I welcome all exchanges/discussions. Hope you are well. Beledarian ***@***.***> 于2026年5月8日周五 13:06写道：

…

@SINAPSA-IC <https://github.com/SINAPSA-IC> You mind sharing your tutorial and/ or post in a discussion here once its done? I'd be interested in your work Thank you for being so engaged :) Do you think this plugin is relevant for the population at large? I mean I personally I do enjoy using and experimenting with local llms in my free time. But for most (non-technical) people it seems like they don't even realize that this is something you can do I'll look into it. Maybe I can try to use an agent to experiment with this. This seems like a refactor they should easily handle. — Reply to this email directly, view it on GitHub <#81 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/CBVVTVI3CEEBB7VSRIFYQTD4ZY43VAVCNFSM6AAAAACYSBZ666VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMOBVGY3DSOA> . You are receiving this because you were mentioned.Message ID: <Beledarian/Beledarians_LM_Studio_Toolbox/repo-discussions/81/comments/16856698 @github.com>

[Beledarian]

@LifetimeVip you have to fully uninstall the old plugin to be able to use the new version. LM Studio does not support updating just by clicking download again. You MUST uninstall first

[SINAPSA-IC]

I say that you or anyone else do not develop plugins for the creators of LM Studio, but for the users of LM Studio, that is people who really need them (just think of it - in the js-code-sandbox plugin that comes with the program, denying the access to the web is the static default, there's no switch for it, and there ain't a no-sandbox version either)

Alrighty. Here's the guide.
However(tm), keep in mind that it was built:

• for a --> quick --> introduction to LM Studio
• for testing the water of LLMs (what's a token, prompt, Top K, etc.), not delving into academic technicalities
• originally in another language; the English version here is the translation courtesy of ChatGPT's basic model

So. it's not a professional, industrial-grade training module, but it makes for an easy lecture for the innocent, curious would-be (or wannabe) user.

20260505-tutorial_lmstudio047_en.pdf

[SINAPSA-IC]

A much more comprehensive guide, to LLMs this time, without mentioning plugins though, was composed by ChatGPT. Gradually ofc. With the guidance of yours truly. And the cover, too.

sinapsa_booklet_llm.pdf

[Beledarian]

I have to say this booklet looks incredibly strong (only skimmed it surface level though)

It's way deeper than I expected.
A few days ago I had an agent swarm create an onboarding book for a codebase and It surprised me in quality. Its around 55k words (it started with 11k which was very insufficient) with a very readable format and has a real human expert vibe. But I have to mention that it built it on top of very detailed existing documentation. So not from scratch

The first one also is well made. It's cool how much effort you put into this :)

[SINAPSA-IC]

Thank you! It was nothing, coloring, indenting, making tables, setting fonts and paragraphs, making the list of internal hrefs, header and footer, rearranging the info in places...
:)
Enjoy!

[SINAPSA-IC]

I've noticed that, when the tools in the list shown in LM Studio are all enabled (their checkbocks checked) - I don't know if this is the default when activating the plugin, or if they are enabled when I click on Allow all tools from this plugin (or whatever wording to that end):

• the initial token count is a bit larger that 8000 after not much of a prompt - this may be a fifth of a 40,000 Context Length ahead of a modest conversation
• the model replies quite sloww, I suspect because it's poring through all respective tools/functions to do something, even when a tool is neither mentioned, nor required

So, if a lightweight version of the plugin would be active - with, say, only the few tools that are expected to be used on a regular basis by a common user:

• web browsing
• fetching content from the web (in general/any website)
• fetching content from Wikipedia
• perform RAG on files and on web pages
• running JavaScript and/or Python for whatever tasks
  the token count would decrease toward (much) less than 8000, and the model would move faster.

The token count would be smaller still, if the functions would be reformed by ChatGPT or a Good code-oriented model that would "optimize and condense/compress" the code of each function down to its smallest size, with the same functionality, and yielding the same results as the initial form.