How to add memory using langgrapg.checkpoint in langgraph_agentic_rag.ipynb file

[SujayC66]

Issue with current documentation:

I want to enhance the current flow by adding memory so that when a user asks about previous information, the system can respond accordingly.
Below is the agentic langgraph flow

Any guidance on this would be greatly appreciated.

Idea or request for content:

No response

[vbarda]

@SujayC66 I would recommend checking these two sections of the documentation: https://langchain-ai.github.io/langgraph/how-tos/persistence/ and https://langchain-ai.github.io/langgraph/concepts/low_level/#threads. if you run into a specific issue with implementation, please let me know!

[SujayC66]

I took the reference of same links to add memory component to my flow but its not working. What's happening here is

when im asking a question it gives response as well as follow up question. But when I ask next question, based on follow up question , using same thread, the flow is getting started from START NODE. Also the question is not getting refined as per previous response

[vbarda]

the flow will always start from the start node on a new graph run -- that's by design. maybe you could paste your code here and i can take a look?

[SujayC66]

workflow = StateGraph(GraphState)

Define the nodes

workflow.add_node("refined_query", refined_query)
workflow.add_node("web_search_exasearch", web_search_exasearch) # web search
workflow.add_node("web_search_bravesearch", web_search_bravesearch)
workflow.add_node("generate", generate) # generate
workflow.add_node("re_refined_query", re_refined_query) # re_refined_query
workflow.add_node("generate_follow_up_questions", generate_follow_up_questions) # generate follow up questions

workflow.set_entry_point("refined_query")
workflow.add_conditional_edges(
"refined_query",
route_question,
{
"web_search_exasearch": "web_search_exasearch",
"web_search_bravesearch": "web_search_bravesearch",
}
)
workflow.add_edge("web_search_exasearch", "generate")
workflow.add_edge("web_search_bravesearch", "generate")
workflow.add_conditional_edges(
"generate",
grade_generation_v_documents_and_question,
{
"useful": "generate_follow_up_questions",
"not useful": "re_refined_query",
},
)
workflow.add_edge("re_refined_query", "refined_query")
workflow.add_edge("generate_follow_up_questions", END)

app = workflow.compile()

total_start_time = time.time()
try:
for output in app.stream(inputs, {"recursion_limit": 20}):
if 'generate' in output and 'generation' in output['generate']:
print("Generation:")
generated_response = output['generate']['generation']
generated_response_1 = output['generate']['stream']
for word in generated_response_1:
print(word, end='', flush=True)
try:
if 'generate_follow_up_questions' in output:
print("Follow-up Question:")
generated_follow_up_question = output['generate_follow_up_questions']['follow_up_question']
web_search_time = output['generate_follow_up_questions']['web_search_elapsed_time']
extracting_urls_time = output['generate_follow_up_questions']['fetch_urls_elapsed_time']
gemini_knowledge_base_time = output['generate_follow_up_questions']['gemini_elapsed_time']
final_response_time = output['generate_follow_up_questions']['generate_elapsed_time']
except:
pass
except:
generated_response = "Could you please ask the question again with more specificity?"
generated_follow_up_question = "I am unable to provide theanswer for your query due to unforeseen reasons, please ask another question"
web_search_time = 0
extracting_urls_time = 0
gemini_knowledge_base_time = 0
final_response_time = 0

total_end_time = time.time()

total_elapsed_time = (total_end_time-total_start_time)*1000