🚀 Get up and running with a local conversational agent in minutes.

This project is a minimal, deep-agentic template designed for experimenting with local LLMs. It combines Chainlit for the chat interface, Ollama for local model inference, and LangGraph for custom agentic workflows.

Use this as a starting point to build your own custom AI personas and agents.

• ⚡ Instant Setup: Clone, install, and chat.
• 🏗️ Template Structure: Clean, modular code ready for customization.
• 💬 Interactive UI: Polished chat interface out-of-the-box.
• 🧠 Extended Thinking: Visualizes the agent's reasoning process using Chainlit steps.
• 🔗 Native LangGraph: Custom graph implementation (Reasoning -> Response) for full control.
• 📊 Telemetry Ready: Integrated with Langfuse for observability.
• 🔒 100% Local: Privacy-first using Llama 3.1 via Ollama.

1. Python 3.10 or higher
2. Ollama running locally.

Install Ollama and pull the Llama 3.1 model:

ollama pull llama3.1

Start the Ollama server:

ollama serve

1. Clone the repository:

   git clone https://github.com/Yn0t-studio/Agent-Launchpad.git
   cd Agent-Launchpad

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

1. Ensure Ollama is running (ollama serve).

2. Run the application:

   chainlit run app.py -w

3. Open http://localhost:8000 to chat.

To enable tracing and observability with Langfuse:

1. Create a .env file in the root directory.

2. Add your Langfuse keys:

   LANGFUSE_PUBLIC_KEY=pk-lf-...
   LANGFUSE_SECRET_KEY=sk-lf-...
   LANGFUSE_HOST=https://cloud.langfuse.com # or your self-hosted instance

3. Restart the application. Telemetry will automatically be enabled if keys are present.

Edit agent.py to configure the agent:

def get_agent():
    # ...
    model = ChatOllama(model="llama3.1", base_url="http://localhost:11434")
    
    # Define custom nodes and workflow
    # See agent.py for the full graph definition
    
    return workflow.compile()