Error adding extra input_variables (ValueError: Missing some input keys: {'history'})

[Erick-ECE]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

`
from langchain.memory import ConversationBufferMemory

def start_chat: Update, context: CallbackContext) -> None:
    
      ...
      
        vectorstore = Chroma(
            embedding_function=embed_model,
            persist_directory="chroma_db_dir",
            collection_name="my_collection"
        )
            
        retriever = vectorstore.as_retriever(search_kwargs={'k': 4})
        custom_prompt_template = """
        Answer using the provided context and the conversation history.
        If the context does not help to answer the question, indicate that you do not have enough information
       
        Context: {context} 
                         
        History: {history}
        Question:
        {question}
        
        [Respond in Spanish]
        """
        prompt = PromptTemplate(template=custom_prompt_template, input_variables=['context','history','question',])

 memory = ConversationBufferMemory(return_messages=True, input_key="query",output_key="result",memory_key="history")

        qa = RetrievalQA.from_chain_type(
            llm=llm,
            chain_type="stuff",
            memory=memory,  # Le pasa la memoria del chat en el prompt
            retriever=retriever,
            return_source_documents=False,
            chain_type_kwargs={"prompt": prompt},
        )

       context.user_data['memory'] = memory

def responder_pregunta(update: Update, context: CallbackContext) -> None:
     ....
     ....
     question = update.message.text

     memory = context.user_data.get('memory')

     memory_variables = memory.load_memory_variables({})

     answer = qa.invoke({"query": pregunta,"history": memory_variables})

     update.message.reply_text(answer['result'])
`

Description

I'm trying to implement chat memory using ConversationBufferMemory, and use the chat history in the prompt to provide context in the awnser of the chat questions
the way I implemented the memory is storing correctly the messages and the answers.

The issue is that the cant pass it to the propt.
I'm trying by using an input_variable History: {history} but have the error : (ValueError: Missing some input keys: {'history'}) even though I'm passing explicity the history: answer= qa.invoke({"query": pregunta,"history": memory_variables})

cold you help me? whats wrong or how this should be done ?

System Info

windows 11
langchain_ollama 0.3.3
LangChain 0.3.26

[onestardao]

Ah yes, the classic LangChain input_variables riddle strikes again — I’ve been there, and the error message is more honest than most humans:

ValueError: Missing some input keys: {'history'}

From your screenshot and code, here’s what’s going on:

---

🚨 Diagnosis:

Even though you're passing "history" manually into qa.invoke(...), LangChain templates still expect all input_variables to match exactly. That means your PromptTemplate must also list "history" — and must be structured exactly as the keys you're passing.

I noticed in your code:

prompt = PromptTemplate(
    template=custom_prompt_template,
    input_variables=["context", "history", "question"]
)

But later when invoking:

answer = qa.invoke({"query": pregunta, "history": memory_variables})

You’re passing query, not question.

➡️ So LangChain is like “hey, I expected question but you gave me query, where’s my snack?”

---

✅ Solution:

Just align the variable names in both places. Either:

Option 1: Match the PromptTemplate

qa.invoke({
    "context": context,
    "history": memory_variables,
    "question": pregunta
})

or

Option 2: Adjust the prompt's input_variables to ["context", "history", "query"]

(But I wouldn’t recommend this unless you want extra confusion.)

---

🧠 Bonus Tip:

You're doing well passing memory manually, but you might want to try using ConversationBufferMemory + Runnable chains, which handle this auto-magically. If you're building something larger — like agentic flows or semantic RAG — we’ve built a system (WFGY engine) that already solves these kinds of prompt-sync issues, especially when you start mixing chain memory + retrieval + user injection.

(And yes — full-text system prompts + drunk mode parsing + memory injection protection — all open source.)

More of these issues (with reproducible examples + workarounds) are mapped here:
🔗 https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md

Hope this helps! Let me know if you need a minimal fix snippet.