Fasttrack· 01· 20 min

Your First LLM Call

What you'll learn

▸Make your first LLM call with LangChain
▸Understand SystemMessage, HumanMessage, and AIMessage
▸Stream tokens in real time

This entire Fasttrack uses LangChain — one library that works with every LLM provider. Learn it once and switch between OpenAI, Anthropic, Google, or local models without rewriting your code.

LLM call flow diagram — A single LLM call: Python sends messages, gets back an AIMessage

Create a model with init_chat_model

LangChain provides a universal factory function called init_chat_model() that can instantiate any supported LLM with a single string. The format is "provider:model-name" — for example, "openai:gpt-4o-mini" or "anthropic:claude-3-haiku-20240307". This means you write your application logic once and swap providers by changing a single string, not by rewriting API calls.

first_call.py

from langchain.chat_models import init_chat_model    # ①
model = init_chat_model('openai:gpt-4o-mini')        # ②
response = model.invoke('What is LangChain?')        # ③
print(response.content)                               # ④

①① Universal factory for any provider — one import covers all of them

②② Model string format: "provider:model-name" — change this one string to swap providers

③③ invoke() sends the message and waits for the full response before returning

④④ response is an AIMessage object — .content gives you the text string

Three message types in LangChain — Three message types control the conversation

Message types: System, Human, AI

LangChain uses three message classes that map directly to the roles in any chat API. SystemMessage sets the model's persona, rules, and constraints — it is invisible to end users but shapes every reply. HumanMessage is the user's input — the question or task. AIMessage is the model's reply and is also what model.invoke() returns. When building multi-turn conversations you maintain a list of these messages and resend them on every call.

message_types.py

from langchain_core.messages import SystemMessage, HumanMessage

messages = [
    SystemMessage(content='You are a concise Python tutor. Use code examples.'),
    HumanMessage(content='What is a list comprehension?'),
]

response = model.invoke(messages)
print(response.content)

Streaming tokens

Instead of waiting for the entire response to be generated before printing anything, model.stream() delivers tokens as they arrive from the model. Each chunk is a partial AIMessage. Printing chunk.content as it arrives creates a responsive, word-by-word experience — the same effect you see in ChatGPT when text appears progressively.

streaming.py

for chunk in model.stream('Write a haiku about Python'):
    print(chunk.content, end='', flush=True)
print()  # newline at the end

Streaming token delivery diagram — Streaming delivers tokens word-by-word for a responsive UX

LLMs are stateless — each invoke() is a fresh call with no memory of previous calls. The model does not remember your name, the task you described last turn, or any prior conversation. You must resend the full message history every time to simulate memory.

Knowledge Check

What type does model.invoke() return?

Recap — what you just learned

✓init_chat_model("provider:model-name") is a universal factory — one function for every provider
✓invoke() waits for the full reply; stream() delivers tokens progressively
✓Three message types: SystemMessage (rules), HumanMessage (user input), AIMessage (model reply)
✓Every invoke() is stateless — the model has zero memory of previous calls

Next up: 02 — Providers: Same Code, Any Model