langgraph/tutorials/rag/langgraph_adaptive_rag/

[giscus[bot]]

langgraph/tutorials/rag/langgraph_adaptive_rag/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/tutorials/rag/langgraph_adaptive_rag/

[whillas-yabble]

Second last example here got stuck in a (seemingly) infinite loop :-/

[JacobGoldenArt]

I would love to see some Memory specific Langgraph tutorials. That explain how to create a conversational agent that can save memories and also recall from previous chat sessions, perhaps using some sort of RAG technique. Similar to you Journaling App. Thanks!

[osasisorae]

Yea, I second this

[osasisorae]

https://langchain-ai.github.io/langgraph/how-tos/persistence/

This could help

[rlancemartin]

also see module 2 LangChain academy, which covers memory in depth -
https://github.com/langchain-ai/langchain-academy/tree/main/module-2

[weissenbacherpwc]

Nice, I really like Langgraph with making apps like this much easier. However on the document grading part I am not that happy. It takes long until the LLM graded all retrieved docs, even if there are only 3 docs with 1000 chars retrieved. What could be done to make the grading part faster?

In a production environment I think it takes too long for me unfortunately.

[ITSAIDI]

It's depeding to Hardware you use a GPU with a good VRAM wokrs good.

[ITSAIDI]

Amazing Tuto, Thanks LangGraph Team

[ITSAIDI]

@LangGraph Team , it's possible to build Multiple Graphs, and then combine them in One System with certain communication between them , so we can handle big a and complexe Tasks ?

[alphaply]

If we put all the nodes into a graph, it will be difficult to manage the state. I think we should turn the workflow into a tool.

[ITSAIDI]

The Idea I found is explained here https://langchain-ai.github.io/langgraph/how-tos/subgraph/

[phuonglv-ltv]

a great tool for multi-agent, Thanks LangGraph Team

[SA203260]

After implementing this code, I get the graph where routing is taking place from the START.

When it reaches to the node web_search, it does not know about the state attribute 'document'. So, it gives an error that it can't append data to a None type.

I am running this on "gpt-4o-mini".

To mitigate this, I added a dummy node starter after START and routing everything from the starter node instead of START.

def starter(state: GraphState):
"""
Instantiate initial Graph states

Args:
    state (dict): The current graph state

Returns:
    state (dict): initial graph state
"""
print("---STARTER---")
question = state["question"]
documents= []

return {"documents": documents, "question": question} 

[githubrahulgupta]

I implemented this code step by step.
Now, while testing with the first run, I get this error:

TypeError: unhashable type: 'RouteQuery'

Please suggest.

[githubrahulgupta]

{
"name": "TypeError",
"message": "unhashable type: 'RouteQuery'",
"stack": "---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[35], line 7
3 # Run
4 inputs = {
5 "question": "What player at the Bears expected to draft first in the 2024 NFL draft?"
6 }
----> 7 for output in app.stream(inputs):
8 for key, value in output.items():
9 # Node
10 pprint(f"Node '{key}':")

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\pregel\init.py:1020, in Pregel.stream(self, input, config, stream_mode, output_keys, interrupt_before, interrupt_after, debug)
1017 break
1019 # panic on failure or timeout
-> 1020 _panic_or_proceed(all_futures, loop.step)
1021 # don't keep futures around in memory longer than needed
1022 del done, inflight, futures

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\pregel\init.py:1450, in _panic_or_proceed(futs, step, timeout_exc_cls)
1448 inflight.pop().cancel()
1449 # raise the exception
-> 1450 raise exc
1452 if inflight:
1453 # if we got here means we timed out
1454 while inflight:
1455 # cancel all pending tasks

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\pregel\executor.py:60, in BackgroundExecutor.done(self, task)
58 def done(self, task: concurrent.futures.Future) -> None:
59 try:
---> 60 task.result()
61 except GraphInterrupt:
62 # This exception is an interruption signal, not an error
63 # so we don't want to re-raise it on exit
64 self.tasks.pop(task)

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\concurrent\futures\_base.py:451, in Future.result(self, timeout)
449 raise CancelledError()
450 elif self._state == FINISHED:
--> 451 return self.__get_result()
453 self._condition.wait(timeout)
455 if self._state in [CANCELLED, CANCELLED_AND_NOTIFIED]:

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\concurrent\futures\_base.py:403, in Future.__get_result(self)
401 if self._exception:
402 try:
--> 403 raise self._exception
404 finally:
405 # Break a reference cycle with the exception in self._exception
406 self = None

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\concurrent\futures\thread.py:58, in _WorkItem.run(self)
55 return
57 try:
---> 58 result = self.fn(*self.args, **self.kwargs)
59 except BaseException as exc:
60 self.future.set_exception(exc)

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\pregel\retry.py:26, in run_with_retry(task, retry_policy)
24 task.writes.clear()
25 # run the task
---> 26 task.proc.invoke(task.input, task.config)
27 # if successful, end
28 break

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langchain_core\runnables\base.py:2878, in RunnableSequence.invoke(self, input, config, **kwargs)
2876 input = context.run(step.invoke, input, config, **kwargs)
2877 else:
-> 2878 input = context.run(step.invoke, input, config)
2879 # finish the root run
2880 except BaseException as e:

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\utils.py:102, in RunnableCallable.invoke(self, input, config, **kwargs)
100 if accepts_config(self.func):
101 kwargs["config"] = config
--> 102 ret = context.run(self.func, input, **kwargs)
103 if isinstance(ret, Runnable) and self.recurse:
104 return ret.invoke(input, config)

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\graph\graph.py:88, in Branch._route(self, input, config, reader, writer)
86 value = input
87 result = self.path.invoke(value, config)
---> 88 return self._finish(writer, input, result)

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\graph\graph.py:115, in Branch._finish(self, writer, input, result)
113 result = [result]
114 if self.ends:
--> 115 destinations = [r if isinstance(r, Send) else self.ends[r] for r in result]
116 else:
117 destinations = result

File c:\Users\Rahul Gupta\anaconda3\envs\myconda\lib\site-packages\langgraph\graph\graph.py:115, in (.0)
113 result = [result]
114 if self.ends:
--> 115 destinations = [r if isinstance(r, Send) else self.ends[r] for r in result]
116 else:
117 destinations = result

TypeError: unhashable type: 'RouteQuery'"
}

[hyyphil]

If want to use DDG, try:

from langchain_community.utilities.duckduckgo_search import DuckDuckGoSearchAPIWrapper
from langchain_core.tools import tool


@tool
def search_engine(query: str):
    """Search engine to the internet."""
    results = DuckDuckGoSearchAPIWrapper()._ddgs_text(query)
    return [{"url": r["href"], "content": r["body"]} for r in results]


web_search_tool = search_engine

[ahammi]

Hi, this code is working very well in my side. However, I have added a memory like this :
memory = MemorySaver()

Compile

compiled_agent = workflow.compile(checkpointer=memory)
config = {"configurable":{"thread_id":"1"}}
output = compiled_agent.stream(input=inputs, config=config)

But it didn't work, any idea why ?

Thanks in advance.
Regards

[ahammi]

I missed to mention that I am calling agent within a Flask App and according to the doc memorysaver and even sqlitesaver are not supported within multithread context. thus I think I will need to implement PostgresSaver. I will let you know about the result, in meanwhile if anyone else did that before (especially agent memory + flask) please do not hesitate to let us know.
Regards

[ahammi]

sorry for the confusion. There is no inference with memorysaver neither sqlitesaver and Flask. In this exemple, The issue was linked to the GraphState, to add the memory there, you will have to change the GrpahState to be inherited from MessageState and change question to messages. also when you invoke the rag you will need to concataenate the whole messages. hoep that helps! working fine for me now !

[jai1007-lab]

If a query is initially routed to web_search because it's unrelated to the index, but later fails validation after generate, it gets redirected to transform_query, which then routes it back to retrieve. This creates an inconsistent logic flow because the original routing decision deemed retrieve irrelevant. The query should instead loop back to web_search or exit gracefully after retries, not switch paths illogically.

Is this a bug or am I tripping?