Phase 1 · The First Chain

In this chapter we run the smallest possible LangChain app and build the mental model of the Model I/O triad.

The Model I/O Triad / Model I/O 三件套

PromptTemplate  ──▶  ChatModel  ──▶  Output
   (input tmpl)        (LLM)          (reply)

• PromptTemplate: fills variables into a fixed prompt template to produce messages.
• ChatModel: actually calls the LLM (via examples/common/llm.py, switchable between OpenAI / DeepSeek / Ollama).
• Output: the model's reply (response.content).

Code Walkthrough / 代码走读

examples/p1/hello_chain.py:

from langchain_core.prompts import ChatPromptTemplate
from examples.common.llm import get_chat_model

def main() -> None:
    model = get_chat_model()
    prompt = ChatPromptTemplate.from_messages([
        ("system", "You are a helpful assistant. Answer in one sentence."),
        ("user", "Explain what {concept} is in one sentence."),
    ])
    chain = prompt | model
    response = chain.invoke({"concept": "StateGraph"})
    print(response.content)

Key points:

• prompt | model is LangChain's LCEL pipe syntax, composing two steps into one runnable unit.
• {concept} is a template variable injected by invoke({"concept": ...}).

Run / 运行

python -m examples.p1.hello_chain

Expected output (example):

A StateGraph is a LangGraph structure that describes an Agent's execution flow using nodes and edges.

Takeaway / 小结

You have run your first Chain. Remember the main line: prompt template → model → output. Every complex Agent later is just more "nodes" and "edges" built on top of it.

Next (Phase 2): combine multiple chains and add memory & retrieval.

Extra Example

Streaming (examples/p1/streaming.py)

Shows both one-shot and streaming calls so you can see the model emit tokens token-by-token.

python -m examples.p1.streaming