LangGraph Agentic AI

A configurable, multi-agent chatbot platform built with LangGraph and Streamlit. Pick an LLM provider and a use case from the sidebar, and the app builds and runs a different agent graph behind the scenes — from a simple chatbot to a tool-using web search agent to a fully autonomous news-fetching pipeline.

Features

• Basic Chatbot — a single-node conversational agent for direct Q&A.
• Chatbot with Web — a tool-using agent that conditionally routes between the LLM and a Tavily search tool, letting the model decide when it needs live web results before answering.
• AI News — an autonomous fetch → summarize → save pipeline that pulls the latest AI news for a chosen time frame (daily/weekly/monthly), summarizes it into Markdown with the LLM, and saves it under AINews/.
• Config-driven UI — LLM providers, models, and use cases are defined in a single .ini file, so the sidebar can be extended without touching UI code.

Requirements

• Python 3.10+
• A Groq API key (free tier available)
• A Tavily API key — only needed for the "Chatbot with web" and "AI News" use cases

Setup

# 1. Clone and enter the project
git clone <this-repo-url>
cd AgenticAI

# 2. Create and activate a virtual environment
python3 -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

Running the app

streamlit run app.py

This starts the app at http://localhost:8501.

Usage

1. Select LLM — currently supports Groq.
2. Select Model — choose a Groq-hosted model (e.g. llama-3.1-8b-instant).
3. Enter your Groq API key — required to initialize the model; entered per-session, never stored on disk.
4. Select usecase:
   • Basic Chatbot — type a message in the chat box at the bottom and press enter.
   • Chatbot with web — also requires a Tavily API key (a field appears once selected). The agent decides on its own whether to search the web before responding.
   • AI News — also requires a Tavily API key. Pick a time frame (Daily/Weekly/Monthly) and click Fetch Latest AI News. The summarized digest is displayed in the app and saved to AINews/<timeframe>_summary.md.

Architecture

Overview

app.py
  └─ main.load_langgraph_agentic_app()
       ├─ LoadStreamlitUI          → renders sidebar, collects user selections
       ├─ GroqLLM                  → initializes the chosen LLM
       ├─ GraphBuilder             → builds a LangGraph StateGraph for the selected use case
       └─ DisplayResultStreamlit   → runs the graph and renders output as chat messages

The app is driven by a single shared State (a TypedDict, see src/langgraphagenticai/state/state.py) that flows through each graph's nodes. Every use case is modeled as its own LangGraph StateGraph, built independently in GraphBuilder, so adding a new agent behavior means adding a new node/edge configuration rather than modifying a shared prompt chain.

Project structure

app.py                                  Entry point — launches the Streamlit app
requirements.txt                        Python dependencies
AINews/                                 Saved AI News summaries (generated at runtime)

src/langgraphagenticai/
├── main.py                             Orchestrates UI → LLM → graph → display
├── llm/
│   └── groqllm.py                      Groq LLM client initialization
├── graph/
│   └── graph_builder.py                Builds the LangGraph StateGraph per use case
├── nodes/
│   ├── basic_chatbot_node.py           Node for the Basic Chatbot graph
│   ├── chatbot_with_Tool_node.py       Node(s) for the web-search chatbot graph
│   └── ai_news_node.py                 Nodes for the AI News fetch/summarize/save graph
├── tools/
│   └── search_tool.py                  Tavily search tool + tool node factory
├── state/
│   └── state.py                        Shared graph State schema
└── ui/
    ├── uiconfig.py                     Reads uiconfig.ini
    ├── uiconfig.ini                    LLM options, models, and use cases (edit to extend)
    └── streamlitui/
        ├── loadui.py                   Renders the sidebar and collects user input
        └── displayresult.py            Runs the graph and renders results in the chat UI

Graphs per use case

Basic Chatbot (graph_builder.basic_chatbot_build_graph)

START → chatbot → END

A single node that invokes the LLM directly on the conversation history.

Chatbot with web (graph_builder.chatbot_with_tools_build_graph)

START → chatbot ⇄ tools
           │
           └─→ END

The LLM is bound to the Tavily search tool. After each chatbot step, tools_condition inspects the model's response: if it requested a tool call, the graph routes to the tools node (which executes the search and returns results to chatbot); otherwise it routes straight to END.

AI News (graph_builder.ai_news_builder_graph)

START → fetch_news → summarize_news → save_result → END

A linear, fully autonomous pipeline:

1. fetch_news — queries Tavily for recent AI news within the selected time range.
2. summarize_news — prompts the LLM to summarize the articles into a structured Markdown digest.
3. save_result — writes the digest to AINews/<timeframe>_summary.md.

Configuration

Sidebar options are driven by src/langgraphagenticai/ui/uiconfig.ini:

[DEFAULT]
PAGE_TITLE = LangGraph: Build Sequetial Agentic AI graph
LLM_OPTIONS = Groq
USECASE_OPTIONS = Basic Chatbot, Chatbot with web, AI News
GROQ_MODEL_OPIONS = llama-3.1-8b-instant, openai/gpt-oss-120b, qwen/qwen3.6-27b, whisper-large-v3-turbo

To add a new model or use case, add it to the relevant comma-separated list — no UI code changes required for the dropdown itself (though a new use case still needs a corresponding graph in graph_builder.py and a branch in displayresult.py).

Notes

• API keys are entered per-session through the UI and are not persisted to disk.
• The AINews/ directory is created automatically on first use of the AI News feature.