Agent stalling and LLM response not returning anything

OSS Product Help
1

I am currently using Langchain for the LLM instances for my agent, however sometimes whenever I run the agent, it does not return anything or takes too long and returns error. I am able to observe this on Langsmith and I just see it spinning. However, I am not really sure what is the cause of the issue. I am using Openrouter as my provider, however every time I call their API normally I am able to get a response all the time, so I am thinking this could be cause by Langchain. Does anyone have any experience with this and what they would recommend? Also, I am thinking of moving off of Langchain for the LLM instances and such, and use Langgrpah with the native APIs, would this be plausible?

2

hi @Art

how do you use OpenRouter through LangChain - via ChatOpenAI with a custom base_url, or via the dedicated langchain-openrouter package?

3

HI @pawel-twardziak ,

Thanks for the quick response. I use it with the init_chat_model using the custom base url. I assume that this is using the ChatOpenAI under the hood to my understanding.

4

hi @Art

When you call init_chat_model with a model name and a custom base_url, the function infers the provider from the model name prefix.

For example:

  • "gpt-4o" → infers "openai" → creates ChatOpenAI
  • "claude-sonnet-4-5" → infers "anthropic" → creates ChatAnthropic

So if you’re doing something like:

from langchain.chat_models import init_chat_model

llm = init_chat_model(
    "gpt-4o",
    base_url="https://openrouter.ai/api/v1",
    api_key="your-openrouter-key",
)

This creates a ChatOpenAI instance with a custom base_url pointing to OpenRouter.

This setup has two critical default behaviors that cause stalling:

Problem 1: No default timeout - ChatOpenAI sets request_timeout to None by default

Problem 2: stream_usage stays disabled with custom base_url - when a custom base_url is set, ChatOpenAI skips enabling stream_usage by default

add explicit timeout and retries:

from langchain.chat_models import init_chat_model

llm = init_chat_model(
    "gpt-4o",
    base_url="https://openrouter.ai/api/v1",
    api_key="your-openrouter-key",
    timeout=60,
    max_retries=2,
    stream_usage=True,
)

or better alternative imho (the package itself is quite old though):

model_provider="openrouter" instead of custom base_url

from langchain.chat_models import init_chat_model

llm = init_chat_model(
    "openai/gpt-4o", 
    model_provider="openrouter",
    timeout=30000, 
    max_retries=1,
)

It requires pip install langchain-openrouter