How to modifiy config to support openrouter api?

[AEONWANG]

Hi,
I thought openrouter api is compatible with langchain_openai just with changing endpoint url so I tried like this:

env file:

OPENAI_API_KEY="openrouter key"
OPENAI_ENDPOINT="https://openrouter.ai/api/v1"
OPENAI_MODEL="qwen/qwq-32b:free"

configuration file:

planner_provider: str = "openai"
planner_model: str = "qwen/qwq-32b:free"
writer_provider: str = "openai"
writer_model: str = "qwen/qwq-32b:free"

fianally it failed with request timeout

appreciate if any suggestion

[aruno14]

AEONWANG I modified the library to use OpenRouter.

My config file is:

thread = {"configurable": {"thread_id": str(uuid.uuid4()),
                           "search_api": "tavily",
                           "planner_provider": "openai",
                           "planner_model": MODEL,
                           "planner_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "writer_provider": "openai",
                           "writer_model": MODEL,
                           "writer_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "max_search_depth": 1,
                           "report_structure": REPORT_STRUCTURE_JP
                           }}

Then I modified open_deep_research/graph.py.
I get the kwargs arguments like below:

planner_kwargs = get_config_value(configurable.planner_kwargs)

And added it to init_chat_model() like below

planner_llm = init_chat_model(model=planner_model, 
                                      model_provider=planner_provider,
                                      **planner_kwargs)

You have to do the same thing for all init_chat_model() call.

It seems ChatOpenAI() does not use OPENAI_ENDPOINT env.

[AEONWANG]

@AEONWANG I modified the library to use OpenRouter.

My config file is:

thread = {"configurable": {"thread_id": str(uuid.uuid4()),
                           "search_api": "tavily",
                           "planner_provider": "openai",
                           "planner_model": MODEL,
                           "planner_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "writer_provider": "openai",
                           "writer_model": MODEL,
                           "writer_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "max_search_depth": 1,
                           "report_structure": REPORT_STRUCTURE_JP
                           }}

Then I modified open_deep_research/graph.py. I get the kwargs arguments like below:

planner_kwargs = get_config_value(configurable.planner_kwargs)

And added it to init_chat_model() like below

planner_llm = init_chat_model(model=planner_model, 
                                      model_provider=planner_provider,
                                      **planner_kwargs)

You have to do the same thing for all init_chat_model() call.

It seems ChatOpenAI() does not use OPENAI_ENDPOINT env.

many thanks, Now I can connect to openrouter but it response like this(I tried model qwen/qwq-32b:free):

ValueError('expected value at line 1 column 1')Traceback (most recent call last):

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\pregel_init_.py", line 2621, in astream
async for _ in runner.atick(

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\pregel\runner.py", line 360, in atick
_panic_or_proceed(

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\pregel\runner.py", line 458, in _panic_or_proceed
raise exc

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\pregel\retry.py", line 128, in arun_with_retry
return await task.proc.ainvoke(task.input, config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\utils\runnable.py", line 643, in ainvoke
input = await step.ainvoke(input, config, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langgraph\utils\runnable.py", line 431, in ainvoke
ret = await asyncio.create_task(coro, context=context)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main./src/open_deep_research/graph.py", line 86, in generate_report_plan
results = structured_llm.invoke([SystemMessage(content=system_instructions_query),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\runnables\base.py", line 3023, in invoke
input = context.run(step.invoke, input, config, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\runnables\base.py", line 5358, in invoke
return self.bound.invoke(
^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 307, in invoke
self.generate_prompt(

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 843, in generate_prompt
return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 683, in generate
self._generate_with_cache(

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 896, in _generate_with_cache
for chunk in self._stream(messages, stop=stop, **kwargs):

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\langchain_openai\chat_models\base.py", line 876, in _stream
for chunk in response:

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\openai\lib\streaming\chat_completions.py", line 72, in iter
for item in self._iterator:

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\openai\lib\streaming\chat_completions.py", line 116, in stream
events_to_fire = self._state.handle_chunk(sse_event)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\openai\lib\streaming\chat_completions.py", line 342, in handle_chunk
self.__current_completion_snapshot = self._accumulate_chunk(chunk)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\motli\Desktop\tool\open_deep_research-main\venv\Lib\site-packages\openai\lib\streaming\chat_completions.py", line 438, in _accumulate_chunk
choice_snapshot.message.parsed = from_json(
^^^^^^^^^^

ValueError: expected value at line 1 column 1

During task with name 'generate_report_plan' and id 'd55f9db7-553a-b1c4-b727-354b9e7250ec'

any suggesion?

[aruno14]

AEONWANG Did you try with another model ? I tried with o3-mini, it works but not with gemini-2.0-flash-001.

[rlancemartin]

Nice! I added this README -
#75 (comment)

[murarduino]

@AEONWANG I modified the library to use OpenRouter.@AEONWANG我修改了库以使用OpenRouter。

My config file is:我的配置文件是：

thread = {"configurable": {"thread_id": str(uuid.uuid4()),
                           "search_api": "tavily",
                           "planner_provider": "openai",
                           "planner_model": MODEL,
                           "planner_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "writer_provider": "openai",
                           "writer_model": MODEL,
                           "writer_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "max_search_depth": 1,
                           "report_structure": REPORT_STRUCTURE_JP
                           }}

Then I modified open_deep_research/graph.py. I get the kwargs arguments like below:我修改了#1。我得到了如下的open_deep_research/graph.py参数：

planner_kwargs = get_config_value(configurable.planner_kwargs)

And added it to init_chat_model() like below并将其添加到init_chat_model() like below

planner_llm = init_chat_model(model=planner_model, 
                                      model_provider=planner_provider,
                                      **planner_kwargs)

You have to do the same thing for all init_chat_model() call.你必须对所有init_chat_model()呼叫做同样的事情。

It seems ChatOpenAI() does not use OPENAI_ENDPOINT env.ChatOpenAI()不使用OPENAI_ENDPOINT env。

Pleasethread = {"configurable"...Where should I write? A file in the legacy/directory? Can I also open a config file?

[xuzihan1]

这个方法

@AEONWANG I modified the library to use OpenRouter.

My config file is:

thread = {"configurable": {"thread_id": str(uuid.uuid4()),
                           "search_api": "tavily",
                           "planner_provider": "openai",
                           "planner_model": MODEL,
                           "planner_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "writer_provider": "openai",
                           "writer_model": MODEL,
                           "writer_kwargs": {"openai_api_base": "https://openrouter.ai/api/v1",
                                             "openai_api_key": os.getenv("OPENROUTER_API_KEY")},
                           "max_search_depth": 1,
                           "report_structure": REPORT_STRUCTURE_JP
                           }}

Then I modified open_deep_research/graph.py. I get the kwargs arguments like below:

planner_kwargs = get_config_value(configurable.planner_kwargs)

And added it to init_chat_model() like below

planner_llm = init_chat_model(model=planner_model, 
                                      model_provider=planner_provider,
                                      **planner_kwargs)

You have to do the same thing for all init_chat_model() call.

It seems ChatOpenAI() does not use OPENAI_ENDPOINT env.

This method is outdated in the current version. I have now implemented a compatible implementation in the new version.

1. In the .env file, add OPENAI_BASE_URL, and make sure it ends with /v1.
2. In the open_deep_research/configuration.py file, modify the names of all models:
   For example, summarizing the model:
   summarization_model: str = Field(
   default="openai:gpt-5-chat-latest",
   metadata={
   "x_oap_ui_config": {
   "type": "text",
   "default": "openai:gpt-5-chat-latest",
   "description": "Model for summarizing research results from Tavily search results"
   }

then all will work