Prompt Hub

What we're building

A ChatPromptTemplate, identical in shape to the ones from the LangChain course, pushed to LangSmith's Prompt Hub under a name, then pulled back down and used to answer a question, proving it round-trips correctly.

What this reveals

Every prompt so far, in this course and the last two, has lived entirely inside a Python file. That works until a prompt needs to be shared across scripts, edited by someone who isn't touching code, or tracked for changes over time, then hardcoding it becomes a liability: which file has the current version, who changed the wording last, was this the version that scored well in Lesson 10?

The Prompt Hub is LangSmith's answer: a prompt stored centrally, under a name, with every push creating a new version you can see and roll back to from the UI. push_prompt stores a ChatPromptTemplate object directly, pull_prompt fetches the current version back as the same kind of object, ready to use in a chain exactly like one you defined locally.

The code, piece by piece

prompt = ChatPromptTemplate.from_messages([
    ("system", "Answer using only the given context, in one short sentence."),
    ("human", "Context:\n{context}\n\nQuestion: {question}"),
])
client.push_prompt(PROMPT_NAME, object=prompt)

Nothing new about building the template itself (LangChain course, Lesson 3). push_prompt is what's new: it serializes the template and stores it in the hub under PROMPT_NAME, creating a new version if the name already exists.

prompt = client.pull_prompt(PROMPT_NAME)

Fetches the current version back, reconstructed as a real ChatPromptTemplate, not just text. From here it's used exactly like any other prompt template, nothing downstream needs to know it came from the hub rather than from local code.

chain = prompt | model
response = chain.invoke({"context": context, "question": question})

Ordinary LCEL, prompt | model, exactly as in the LangChain course.