🤖 Autonomous Local AI Chatbot Orchestrator

A privacy-first, fully offline conversational AI assistant and RAG (Retrieval-Augmented Generation) agent built in Python using LangGraph, LangChain, Ollama, and ChromaDB.

This application runs entirely on local compute, allowing users to query private documents (PDFs, spreadsheets) and execute agentic tool workflows without transmitting data to external third-party cloud APIs.

---

🚀 Key Features

• Offline Local LLM: Integrates with local language models (such as mistral, llama3, phi3) executed via Ollama, eliminating external API costs and security concerns.
• LangGraph-Driven State Management: Orchestrates conversations using a finite-state machine graph to handle conditional routing, memory tracking, and agentic workflows.
• Function-Calling & Tool Execution: Includes conditional node routing to resolve specific requests (e.g., date/time retrieval, live weather updates, spreadsheet lookups) prior to passing contexts to the LLM.
• Document Retrieval Pipeline (RAG): Pre-equipped with integrations for PyPDF, Pandas/OpenPyXL, SentenceTransformers, and Chroma vector databases to ingest, chunk, embed, and index private files for contextual search.
• Memory-Preserved Multi-Turn Chat: Implements full message history tracking within the graph state to ensure context is retained across conversational turns.

---

🧠 Workflow Architecture

The application routes conversational inputs based on detected intent:

graph TD
    A[User Input] --> B[LangGraph Entry Point]
    B --> C{Intent Classifier / Tool Router}
    C -- Time / Date query --> D[Time Tool Node]
    C -- Weather query --> E[Weather API Tool Node]
    C -- RAG Document Query --> F[Chroma Vector Search Node]
    C -- Standard Chat --> G[Local LLM Node via Ollama]
    D --> H[State Update: Append Message]
    E --> H
    F --> G
    G --> H
    H --> I[Output to CLI / Interface]
    I --> J[End Node]

Loading

---

🛠️ Technology Stack

• Orchestration: LangGraph (StateGraph architecture)
• Chaining & Prompts: LangChain, LangChain-Community
• Inference Engine: Ollama (Default local model: mistral)
• Vector Database: ChromaDB
• Embeddings: SentenceTransformers (Local HuggingFace embeddings)
• Document Parsers: PyPDF (for PDFs), Pandas & OpenPyXL (for XLSX datasets)

---

⚙️ Setup and Installation

1. Prerequisites

• Python 3.10 or higher
• Ollama installed locally

2. Download and Start the Local LLM

Install Ollama and pull your model of choice (e.g., Mistral):

# Pull the default model
ollama pull mistral

Ensure that the Ollama service is running on your background.

3. Install Dependencies

pip install -r requirements.txt

4. Running the Chatbot

Launch the terminal-based interactive agent:

python app.py

---

🛡️ License

Distributed under the MIT License. See LICENSE for more information.