01_basic_chatbot

Step 01: Basic Chatbot

The simplest possible starting point — call an LLM and have a multi-turn conversation.

What you'll learn

• How to use the OpenAI SDK with an OpenAI-compatible provider (InfiniAI)
• How to maintain multi-turn conversation history via a messages list
• How to configure system prompts

Architecture

User Input ──> messages list ──> LLM API ──> Assistant Response
                   ▲                              │
                   └──────────────────────────────┘
                        (append to history)

The key data structure is the messages list — an ordered sequence of role-tagged messages (system, user, assistant) that represents the full conversation context sent to the LLM on every call.

Setup

pip install -r requirements.txt
export INFINI_API_KEY="your-api-key-here"

The default base URL points to InfiniAI (https://cloud.infini-ai.com/maas/v1). Override with:

export INFINI_BASE_URL="https://your-provider.com/v1"

Usage

List available models:

python chatbot.py --list-models

Start chatting (default model: deepseek-v3):

python chatbot.py

Use a specific model:

python chatbot.py --model qwen3-32b

Key Concepts

The Messages List

Every LLM API call is stateless — the model has no memory. Multi-turn conversation works by sending the entire conversation history each time:

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is Python?"},
    {"role": "assistant", "content": "Python is a programming language..."},
    {"role": "user", "content": "How do I install it?"},  # new turn
]

This is the fundamental pattern that all LLM applications build upon. In later steps, we'll need to manage this list carefully as conversations grow long.

OpenAI-Compatible API

The OpenAI SDK can talk to any provider that implements the same API format. We just change base_url:

client = OpenAI(api_key="...", base_url="https://cloud.infini-ai.com/maas/v1")

This means our code works with OpenAI, InfiniAI, Ollama, vLLM, and many others.

What's Next

In Step 02, we'll add token counting and context window management — essential infrastructure before we start building agent capabilities.