Documentation

1. Getting Started

NGIBS is designed to act as an orchestrator for local LLMs. To function, it requires a local inference server to talk to. We strictly use and support Ollama.

Prerequisites

• Windows 10/11, macOS, or Linux
• Ollama installed and running locally
• At least 8GB of RAM (16GB+ recommended for 7B+ parameter models)

Installation (Pre-built Binaries)

If you downloaded the `.exe` (Windows) or `.AppImage` (Linux) release:

1. Double click the executable.
2. On first launch, NGIBS will verify the connection to Ollama and initialize the local ChromaDB vector storage in your AppData/Home directory.

Installation (From Source)

For developers who want to run NGIBS via Python directly:

git clone https://github.com/avarshvir/ngibs.git
cd ngibs
python -m venv venv

# Activate virtual environment
# On Windows:
venv\Scripts\activate
# On Linux/Mac:
source venv/bin/activate

pip install -r requirements.txt
python main.py

---

2. The Four Core Modes

NGIBS uses a dynamic reasoning router. Depending on which button you toggle in the chat interface, the system alters its operational flow entirely.

Quick Search

Behavior: Direct LLM Inference.
When toggled, your prompt is sent directly to the local LLM along with the current session's memory. This is completely offline and the fastest way to get general knowledge or coding assistance.

Live Search

Behavior: Single-pass Web Scraping + Inference.
The system intercepts your prompt and decides if external data is needed. If yes, it pings the DuckDuckGo API, pulls URLs, and scrapes the raw text using BeautifulSoup. This text is injected into the LLM's context window, forcing the model to answer based only on the newly retrieved facts. Sources are appended as citations.

Deep Search

Behavior: Agentic Recursive Reasoning.
The most powerful feature. NGIBS breaks your prompt into sub-queries. It searches the web, reads pages, evaluates if it has enough information, and if not, searches again. Once it gathers enough verified data, it compiles a structured report and saves it locally as a .md, .pdf, or .docx file.

Context Aware

Behavior: Long-term Vector Memory.
This mode leverages ChromaDB. Instead of just looking at the last 10 messages, NGIBS embeds your current prompt into a vector, searches your entire historical database for relevant past conversations or uploaded files, and brings them into the active context window.

---

3. Local File Attachments (RAG)

You can drag and drop files directly into the NGIBS chat input. This initiates a local Retrieval-Augmented Generation (RAG) workflow.

Supported Formats

• PDFs (.pdf)
• Word Documents (.docx)
• Plain Text & Code (.txt, .py, .cpp, .js, etc.)
• CSV Data (.csv)

When a file is attached, NGIBS uses LangChain to chunk the text, embeds it using a lightweight local embedding model, and stores it in your ChromaDB. You can then interrogate the document in "Quick Search" or "Context Aware" mode without the file ever leaving your PC.

---

4. LLM & Memory Management

Unlike cloud platforms, you are the administrator of your AI.

Managing Models

Click the Settings (Gear) Icon in the sidebar to open the LLM Manager. From here, you can:

• Pull Models: Type the name of an Ollama registry model (e.g., phi3, mistral-nemo) and click Pull. NGIBS will download it directly via the Ollama API.
• Delete Models: Free up hard drive space by deleting unused models.
• Swap Models: Use the dropdown at the top of the chat window to switch models mid-conversation (e.g., use a 3B model for a simple question, then swap to an 8B model for code generation).

System Tone (System Prompts)

In the Settings menu, you can alter the fundamental personality of the AI. By changing the "System Tone", you rewrite the invisible base prompt the LLM receives. Want the AI to act like a Senior Python Developer who gives concise answers? Set it here.

Memory Management

Under the "Memory" tab in settings, you can wipe individual chat spaces, clear short-term memory, or permanently flush the ChromaDB vector database if you need to strictly sanitize your environment.