Technical Reference

Keyboard Shortcuts

Navigate the PyQt6 interface without taking your hands off the keyboard.

Shortcut	Action
`Ctrl` + `K`	Focus chat input box
`Ctrl` + `/`	Cycle through Search Modes (Quick -> Live -> Deep -> Context)
`Ctrl` + `N`	Create a New Chat Space
`Ctrl` + `B`	Toggle left sidebar (Show/Hide Chat History)
`Ctrl` + `Shift` + `S`	Open Settings panel
`Esc`	Cancel active Deep Search loop or generation

CLI Flags

If you are launching NGIBS from the terminal or modifying the executable shortcut, you can pass the following arguments:

--debug : Enables verbose LangChain output and HTTP tracing in the console.
--host [URL] : Override the default Ollama API address (e.g., --host http://192.168.1.100:11434 for a networked inference server).
--reset-db : Warning. Wipes the local ChromaDB vector store upon initialization. Use this if your context memory becomes corrupted.

Configuration File

NGIBS stores its persistent settings in a JSON configuration file. You can edit this file manually while the application is closed.

{
  "ollama_host": "http://localhost:11434",
  "default_model": "mistral-nemo",
  "max_recursive_depth": 3,
  "theme": "dark",
  "system_prompt": "You are NGIBS. Provide concise, highly accurate, and cited answers.",
  "auto_save_exports": true
}

Config Keys Explained

max_recursive_depth: Defines how many times a Deep Search loop is allowed to fail/retry before forcing a final summary. Maximum recommended is 5.
system_prompt: The invisible base instructions fed to the LLM at the start of every chat space.

Local Directory Paths

Want to manually backup your data or wipe the slate clean? Here is where NGIBS stores your files based on your OS:

Windows: %APPDATA%\NGIBS\
macOS: ~/Library/Application Support/NGIBS/
Linux: ~/.config/ngibs/

Inside this directory, you will find:

/chroma_db/ - The vector storage for your uploaded files and context.
/chats/ - SQLite database containing your message histories.
/exports/ - Default save location for generated reports.

Supported Export Formats

When using Deep Search, you can command the AI to export its findings natively. NGIBS parses the output through Python rendering libraries to generate clean files.

How to trigger: Simply append "Export this as a PDF" or "Save as a Word Doc" to the end of your Deep Search prompt.

Markdown (.md): Raw, unstyled text. Best for copying into GitHub or Notion.
PDF (.pdf): Rendered via ReportLab. Includes formatting, bold text, and clickable citation links.
Word (.docx): Rendered via python-docx. Best for editing the report later.