Key Features

File System Mastery

• Full Control: Create, read, update, delete, move, and copy files.
• Safe & Secure: All operations are sandboxed to your workspace directory to prevent path traversal attacks.
• Smart Updates: Use replace_text_in_file to make surgical edits instead of rewriting large files.
• Batch Processing: save_file supports creating multiple files in one go.
• Cleanup: Use delete_files_by_pattern to wipe temporary files instantly.

Encountering issues? Feel free to submit them on GitHub.

Find this project helpful? Consider giving it a ⭐ on GitHub or contributing! Thank you for using the toolbox.

Recent Updates (v1.3.2)

• 🛠️ Tool Reordering & Optimization: Reordered the tools list to prioritize casual utilities, improving agent tool selection. Upgraded the web search tool with smart Chrome detection and fallback.
• 🤖 Sub-Agent Reliability Rework: Overhauled the sub-agent loop to prevent infinite loops, improved tool call parsing, unified path/content normalization, and added explicit task completion/abortion capabilities (TASK_FAILED).
• ✨ New Sub-Agent Tools: Empowered sub-agents with multi_replace_text, search_directory, and background command execution. Main agents now support robust batch file saving.

Autonomous Agents

• Secondary Agent: Delegate complex tasks (coding, summarization) to a second local model/server with support for the main model already used and loaded by LM Studio!
• Auto-Save: When the sub-agent generates code, the system automatically detects and saves it to your disk. No more copy-pasting!
• Auto-Debug: (Optional) Triggers a "Reviewer" agent to analyze generated code and fix bugs automatically before returning the result.
• Structured Handoff: Sub-agents can return a dedicated handoff_message for the main agent to relay findings/research.
• Project Context: Agents can read beledarian_info.md to understand your project's history.

Code Execution

• Sandboxed: Run JavaScript (Deno) and Python code.
• Terminal: Execute shell commands or open real terminal windows for interactive tasks.

[!WARNING] Enabling shell or terminal execution allows the model to run arbitrary commands on your system. If enabled, the model may be able to modify files and escape the sandbox environment.

Web & RAG

• Research: Search DuckDuckGo, Wikipedia, or fetch raw web content.
• Advanced Browser Navigation: Persistent browser_session_open/control/close flow for multi-step browsing and automation.
• Web RAG: Chat with website content.
• Local RAG: Semantic search over your workspace files (rag_local_files).

Requirements

• Node.js (v18+)
• LM Studio (v0.3.0+)

💡 Tip: Need persistent long-term memory for your agent? Check out my other project: Local Memory MCP – A privacy-first memory server with knowledge graph support.

Installation

The plugin can be installed using the following link:

https://lmstudio.ai/beledarian/beledarians-lm-studio-tools

Alternatively, you can install it manually for development purposes.

Development

If you want to contribute to the development of this plugin, you can follow these steps:

• Clone the repository:

  git clone https://github.com/Beledarian/Beledarians_LM_Studio_Toolbox.git
  cd Beledarians_LM_Studio_Toolbox
  

• Install dependencies:

• Run in development mode: From within the project directory, run the following command:

  This will start the plugin in development mode. LM Studio should automatically pick it up. Any changes you make to the source code will cause the plugin to automatically reload.

Configuration

Access these settings in the LM Studio "Plugins" tab:

• Enable Secondary Agent: Unlock the power of sub-agents.
• Sub-Agent Profiles: Custom prompts for "Coder", "Reviewer", etc.
• Auto-Debug Mode: Automatically review sub-agent code.
• Sub-Agent Debug Logging: Toggle detailed sub-agent parsing logs for troubleshooting.
• Sub-Agent Auto-Save: Toggle automatic file saving (Default: On).
• Show Full Code Output: Toggle whether to display the full code in chat or hide it for brevity (files are still saved).
• Default Workspace Path: Set the startup workspace directory used by the plugin.
• Safety: Enable/Disable "Allow Code Execution" for Python/JS/Shell.
• Browser Safety: Browser automation for sub-agents requires all three toggles: Allow Browser Control + Sub-Agent: Allow Web Search + Sub-Agent: Allow Browser Control.

Available Tools

File System

• list_directory, change_directory, make_directory
• read_file, save_file (supports batch), delete_path
• replace_text_in_file: Precision editing.
• delete_files_by_pattern: Regex-based cleanup.
• move_file, copy_file, find_files, get_file_metadata
• fuzzy_find_local_files: Levenshtein-based fuzzy file path/name search.

Agent

• consult_secondary_agent: The powerhouse tool. Delegates tasks, handles file creation, and manages sub-agent loops.

Web

• web_search (DuckDuckGo API + HTML fetch/browser fallback), wikipedia_search
• fetch_web_content, rag_web_content
• browser_session_open, browser_session_control, browser_session_close (persistent page automation; preferred for multi-step navigation, with deduped page-text output unless full_read=true)
• browser_open_page (stateless one-shot Puppeteer read)
• Workaround tip: if selector-based click fails, use an evaluate action to click by text, then call with full_read=true.

{
  "actions": [
    {
      "type": "evaluate",
      "script": "const link=[...document.querySelectorAll('a')].find(a=>a.textContent?.includes('LICENSE')); if(link) link.click();"
    }
  ],
  "full_read": true
}

Execution

• run_javascript, run_python
• execute_command (Background), run_in_terminal (Interactive)

Utils

• rag_local_files: Search your code.
• save_memory: Long-term memory.
• get_system_info, read_clipboard, write_clipboard

Developer Guide

See CODE_OVERVIEW.md for architectural details.

Key Features

File System Mastery

• Full Control: Create, read, update, delete, move, and copy files.
• Safe & Secure: All operations are sandboxed to your workspace directory to prevent path traversal attacks.
• Smart Updates: Use replace_text_in_file to make surgical edits instead of rewriting large files.
• Batch Processing: save_file supports creating multiple files in one go.
• Cleanup: Use delete_files_by_pattern to wipe temporary files instantly.

Encountering issues? Feel free to submit them on GitHub.

Find this project helpful? Consider giving it a ⭐ on GitHub or contributing! Thank you for using the toolbox.

Recent Updates (v1.3.2)

• 🛠️ Tool Reordering & Optimization: Reordered the tools list to prioritize casual utilities, improving agent tool selection. Upgraded the web search tool with smart Chrome detection and fallback.
• 🤖 Sub-Agent Reliability Rework: Overhauled the sub-agent loop to prevent infinite loops, improved tool call parsing, unified path/content normalization, and added explicit task completion/abortion capabilities (TASK_FAILED).
• ✨ New Sub-Agent Tools: Empowered sub-agents with multi_replace_text, search_directory, and background command execution. Main agents now support robust batch file saving.

Autonomous Agents

• Secondary Agent: Delegate complex tasks (coding, summarization) to a second local model/server with support for the main model already used and loaded by LM Studio!
• Auto-Save: When the sub-agent generates code, the system automatically detects and saves it to your disk. No more copy-pasting!
• Auto-Debug: (Optional) Triggers a "Reviewer" agent to analyze generated code and fix bugs automatically before returning the result.
• Structured Handoff: Sub-agents can return a dedicated handoff_message for the main agent to relay findings/research.
• Project Context: Agents can read beledarian_info.md to understand your project's history.

Code Execution

• Sandboxed: Run JavaScript (Deno) and Python code.
• Terminal: Execute shell commands or open real terminal windows for interactive tasks.

[!WARNING] Enabling shell or terminal execution allows the model to run arbitrary commands on your system. If enabled, the model may be able to modify files and escape the sandbox environment.

Web & RAG

• Research: Search DuckDuckGo, Wikipedia, or fetch raw web content.
• Advanced Browser Navigation: Persistent browser_session_open/control/close flow for multi-step browsing and automation.
• Web RAG: Chat with website content.
• Local RAG: Semantic search over your workspace files (rag_local_files).

Requirements

• Node.js (v18+)
• LM Studio (v0.3.0+)

💡 Tip: Need persistent long-term memory for your agent? Check out my other project: Local Memory MCP – A privacy-first memory server with knowledge graph support.

Installation

The plugin can be installed using the following link:

https://lmstudio.ai/beledarian/beledarians-lm-studio-tools

Alternatively, you can install it manually for development purposes.

Development

If you want to contribute to the development of this plugin, you can follow these steps:

• Clone the repository:

  git clone https://github.com/Beledarian/Beledarians_LM_Studio_Toolbox.git
  cd Beledarians_LM_Studio_Toolbox
  

• Install dependencies:

• Run in development mode: From within the project directory, run the following command:

  This will start the plugin in development mode. LM Studio should automatically pick it up. Any changes you make to the source code will cause the plugin to automatically reload.

Configuration

Access these settings in the LM Studio "Plugins" tab:

• Enable Secondary Agent: Unlock the power of sub-agents.
• Sub-Agent Profiles: Custom prompts for "Coder", "Reviewer", etc.
• Auto-Debug Mode: Automatically review sub-agent code.
• Sub-Agent Debug Logging: Toggle detailed sub-agent parsing logs for troubleshooting.
• Sub-Agent Auto-Save: Toggle automatic file saving (Default: On).
• Show Full Code Output: Toggle whether to display the full code in chat or hide it for brevity (files are still saved).
• Default Workspace Path: Set the startup workspace directory used by the plugin.
• Safety: Enable/Disable "Allow Code Execution" for Python/JS/Shell.
• Browser Safety: Browser automation for sub-agents requires all three toggles: Allow Browser Control + Sub-Agent: Allow Web Search + Sub-Agent: Allow Browser Control.

Available Tools

File System

• list_directory, change_directory, make_directory
• read_file, save_file (supports batch), delete_path
• replace_text_in_file: Precision editing.
• delete_files_by_pattern: Regex-based cleanup.
• move_file, copy_file, find_files, get_file_metadata
• fuzzy_find_local_files: Levenshtein-based fuzzy file path/name search.

Agent

• consult_secondary_agent: The powerhouse tool. Delegates tasks, handles file creation, and manages sub-agent loops.

Web

• web_search (DuckDuckGo API + HTML fetch/browser fallback), wikipedia_search
• fetch_web_content, rag_web_content
• browser_session_open, browser_session_control, browser_session_close (persistent page automation; preferred for multi-step navigation, with deduped page-text output unless full_read=true)
• browser_open_page (stateless one-shot Puppeteer read)
• Workaround tip: if selector-based click fails, use an evaluate action to click by text, then call with full_read=true.

{
  "actions": [
    {
      "type": "evaluate",
      "script": "const link=[...document.querySelectorAll('a')].find(a=>a.textContent?.includes('LICENSE')); if(link) link.click();"
    }
  ],
  "full_read": true
}

Execution

• run_javascript, run_python
• execute_command (Background), run_in_terminal (Interactive)

Utils

• rag_local_files: Search your code.
• save_memory: Long-term memory.
• get_system_info, read_clipboard, write_clipboard

Developer Guide

See CODE_OVERVIEW.md for architectural details.