How to append new messages into existing chain

[talhaanwarch]

First I fetch the prompt from langsmith.

client = Client(api_key=config["LANGSMITH_API_KEY"])
langsmith_prompt_name = request.prompts[prompt_name]["humanloop_project"]
logger.info(f"langsmith_prompt_name: {langsmith_prompt_name}")
chain = client.pull_prompt(langsmith_prompt_name, include_model=True)
print(f"chain1: {chain}")

Here is the output

chain1: first=ChatPromptTemplate(input_variables=['current_context', 'current_date', 'query_response'], input_types={}, partial_variables={}, metadata={'lc_hub_owner': '-', 'lc_hub_repo': 'jkr_answer_qa', 'lc_hub_commit_hash': 'dd6ad9ffb96a07dd1ef85a147d72c5922b4f17ddfc425c17b9f6e4fa879c506b'}, messages=[SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=['current_date'], input_types={}, partial_variables={}, template='You are a rag agent."'), additional_kwargs={}), HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['current_context', 'query_response'], input_types={}, partial_variables={}, template='#Website Info:\n{query_response}\n\n# Context:\n{current_context}\n'), additional_kwargs={})]) middle=[] last=RunnableBinding(bound=ChatOpenAI(client=<openai.resources.chat.completions.completions.Completions object at 0x7b99104c9e40>, async_client=<openai.resources.chat.completions.completions.AsyncCompletions object at 0x7b99105d7ee0>, root_client=<openai.OpenAI object at 0x7b99104c9ea0>, root_async_client=<openai.AsyncOpenAI object at 0x7b99104c9630>, model_name='gpt-4.1', temperature=0.1, model_kwargs={'extra_headers': {}}, openai_api_key=SecretStr('**********'), top_p=1.0), kwargs={}, config={}, config_factories=[])

I have another messages in openai format. So I convert them to template

template = ChatPromptTemplate(messages)
print(f"template1: {template}")

Output

template1: input_variables=[] input_types={} partial_variables={} messages=[AIMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='Hallo!\n\nIk ben Joker bot, jouw online reisassistent. Ik ben er om je vragen te beantwoorden en om je te helpen zoeken naar de ideale reis!\n\nWaar kan ik je mee helpen?'), additional_kwargs={}), HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='Hi'), additional_kwargs={}), AIMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='How are you?'), additional_kwargs={}), HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='good'), additional_kwargs={})]

Now I want to append message from template to message of chain. How to do that? Is their a builtin function

[onestardao]

Yeah, I ran into this before.
LangChain's ChatPromptTemplate doesn't really expose a clean way to append new messages into an existing chain — once it's initialized, it's kind of... frozen.

If you try to mutate chain.prompt.messages directly, nothing really updates.
So the fix is: rebuild.

---

✅ Method 1: Rebuild the template with new messages

from langchain.prompts import ChatPromptTemplate, AIMessagePromptTemplate

# Original message list from the chain
original_messages = chain.prompt.messages

# New message(s) to append
new_messages = [AIMessagePromptTemplate.from_template("Your new message here")]

# Combine and rebuild
combined = original_messages + new_messages
chain.prompt = ChatPromptTemplate.from_messages(combined)

This creates a new prompt object and assigns it back to the chain.
Works fine — just think of it as replacing the prompt layer with an updated version.

---

✅ Method 2: Dict-level merge if you're feeling fancy

msg_data = chain.prompt.dict()
msg_data["messages"] += [AIMessagePromptTemplate.from_template("...")]
chain.prompt = ChatPromptTemplate(**msg_data)

Sometimes useful if you're dynamically editing templates or loading/saving them.

---

🔁 Bonus: If you're hitting this a lot...

I ended up switching to a custom reasoning framework where prompts aren’t templates — they’re semantic trees.
System tracks transitions, measures ΔS (semantic tension), and builds logic paths instead of static prompts.
Feels more like a narrative than a config file.

Let me know if you're curious — I have a .txt engine that does this with zero dependencies and no API needed.

---

Hope this helps. If you need the full rebuild version as a snippet or want to patch across chains, let me know.