How to use OpenRouter API for DeepSeek?

[agn-7]

Bug Description

I'm trying to use DeepSeek model on Browser-use, however, I don't have a direct API key from Deepseek, instead, I have an API key from Open Router that supports all LLM models. Now, based on the Open Router API doc and the browser-use doc regarding customizing the LLM model I have reached the following code. However, the browser-use opens up a browser instance, but nothing happens and the page remains empty!

Reproduction Steps

Here are the corresponding logs:

INFO     [browser_use] BrowserUse logging setup complete with level info
INFO     [root] Anonymized telemetry enabled. See https://docs.browser-use.com/development/telemetry for more information.
INFO     [agent] 🚀 Starting task: 1. Go to https://www.reddit.com/r/LocalLLaMA 2. Search for 'browser use' in the search bar3. Click on first result4. Return the first comment
INFO     [agent] 📍 Step 1
ERROR    [agent] ❌ Result failed 1/3 times:
 Could not parse response.
INFO     [agent] 📍 Step 1
ERROR    [agent] ❌ Result failed 2/3 times:
 {'message': 'Provider returned error', 'code': 429, 'metadata': {'raw': '{\n  "error_msg": "Authentication Fails. Multiple 401 errors detected. Please wait for 1 minute before trying again."\n}', 'provider_name': 'DeepSeek'}}
INFO     [agent] 📍 Step 1
ERROR    [agent] ❌ Result failed 3/3 times:
 {'message': 'Provider returned error', 'code': 429, 'metadata': {'raw': '{\n  "error_msg": "Authentication Fails. Multiple 401 errors detected. Please wait for 1 minute before trying again."\n}', 'provider_name': 'DeepSeek'}}
ERROR    [agent] ❌ Stopping due to 3 consecutive failures
INFO     [agent] Created GIF at agent_history.gif

Code Sample

Here is the code:

import asyncio
import os

from dotenv import load_dotenv
from langchain_openai import ChatOpenAI

from browser_use import Agent

# dotenv
load_dotenv()

api_key = "sk-or-v1-xxx"
if not api_key:
	raise ValueError('DEEPSEEK_API_KEY is not set')


async def run_search():
	agent = Agent(
		task=(
			'1. Go to https://www.reddit.com/r/LocalLLaMA '
			"2. Search for 'browser use' in the search bar"
			'3. Click on first result'
			'4. Return the first comment'
		),
		llm=ChatOpenAI(
			base_url='https://openrouter.ai/api/v1',
			model='deepseek/deepseek-chat',
			api_key=api_key,
		),
		use_vision=False,
	)

	await agent.run()


if __name__ == '__main__':
	asyncio.run(run_search())

Version

0.1.35

LLM Model

DeepSeek Coder

Operating System

Ubuntu 22.04

[techlism]

I am facing the same issue. Any help would be appreciated.

[prigozhinfan]

I am facing a simialr issue, with both openai/chatgpt-4o-latest and deepseek/deepseek-r1:

Error code: 404 - {'error': {'message': 'No endpoints found that support tool use. To learn more about provider routing, visit: https://openrouter.ai/docs/provider-routing', 'code': 404}}

the deepseek/deepseek-chat was working yesterday, but not working today with this:

ERROR    [agent] ❌ Result failed 1/3 times:
 Could not parse response.

[kadavilrahul]

I think gemini works well and is free. There is no advantage of other models it seems.

[agn-7]

@kadavilrahul

• Is the Gemini API key really free?
• What is your opinion between gpt-4o and Gemini?
• does Gemini support vision?

[kadavilrahul]

Gemini is totally free. You don't need to think much about the model because this script only scrapes and processes data. Any lite weight model works well. I have made a readymade script with terminal inputs for gemini. You may check it.
https://github.com/kadavilrahul/browser-use-shell

[machinedev]

Hi, is there any solution with this? I am facing the same issue here, is it possible using another API endpoint? like deepinfra, openrouter, etc. Cause I can't find any reference with the supported models here: https://docs.browser-use.com/customize/supported-models

[kazGuido]

came with the same problem haha, i cant use it from China without extra gymnastics so trying to go the openrouter route, have you found anything ?

Hi, is there any solution with this? I am facing the same issue here, is it possible using another API endpoint? like deepinfra, openrouter, etc. Cause I can't find any reference with the supported models here: https://docs.browser-use.com/customize/supported-models

[firstcomeuropeag]

same here.

[kazGuido]

same here.
my findings:

• disable the usage of vision ( for some reason it doesnt work with deepseek, but i tried gemini flask 2.0 and it worked )
• use another model from Openrouter

[machinedev]

same here.
my findings:

• disable the usage of vision ( for some reason it doesnt work with deepseek, but i tried gemini flask 2.0 and it worked )
• use another model from Openrouter

thanks mate @kazGuido, will try it

[agn-7]

@kazGuido can you share your code regarding the integration of Openrouter and browser-use to use gemini 2.0 flash? I tested it with this model and I got the same error that I get with the deepseek model!

[UnixSafe]

I had the same problem and eventually found this workaround

Required Modifications

To integrate OpenRouter with browser-use, several modifications are necessary:

1. Installing Dependencies

pip install langchain-community

2. Configuring the Model in Your Main Script

Replace the import and initialization of the LLM model with OpenRouter integration via the OpenAI-compatible API:

# Replace this:
from langchain_google_genai import ChatGoogleGenerativeAI

# With this:
from langchain_community.chat_models import ChatOpenAI

# Then, instead of this:
llm = ChatGoogleGenerativeAI(
    model='gemini-2.0-flash-exp',
    api_key=SecretStr(api_key),
    temperature=0.0
)

# Use this:
llm = ChatOpenAI(
    model="google/gemini-2.0-pro-exp-02-05:free",  # Or any other model available on OpenRouter
    openai_api_key=openrouter_api_key,
    openai_api_base="https://openrouter.ai/api/v1",
    temperature=0.0,
    max_tokens=4096,
    model_kwargs={
        "tool_choice": "auto"  # Allows the model to decide when to use tools
    }
)

3. Modifying Message Handling in browser-use

For browser-use to work correctly with OpenRouter, you need to modify how messages are processed. Here are the changes to make to the agent/message_manager/utils.py file:

def convert_input_messages(input_messages: list[BaseMessage], model_name: Optional[str]) -> list[BaseMessage]:
    """Convert input messages to a format that is compatible with the planner model"""
    if model_name is None:
        return input_messages
    # Handle deepseek models
    if model_name == 'deepseek-reasoner' or model_name.startswith('deepseek-r1'):
        converted_input_messages = _convert_messages_for_non_function_calling_models(input_messages)
        merged_input_messages = _merge_successive_messages(converted_input_messages, HumanMessage)
        merged_input_messages = _merge_successive_messages(merged_input_messages, AIMessage)
        return merged_input_messages
    # Handle OpenRouter/Gemini models
    if 'gemini' in str(model_name).lower() or str(model_name).startswith('google/'):
        logger.info(f"Converting messages for OpenRouter/Gemini model: {model_name}")
        converted_input_messages = _convert_messages_for_non_function_calling_models(input_messages)
        return converted_input_messages
    return input_messages

4. Modifying the get_next_action Method in the Agent Class

To avoid errors related to function calling, modify the get_next_action method in the agent/service.py file:

@time_execution_async('--get_next_action (agent)')
async def get_next_action(self, input_messages: list[BaseMessage]) -> AgentOutput:
    """Get next action from LLM based on current state"""
    input_messages = self._convert_input_messages(input_messages)

    # Special handling for OpenRouter/Gemini models
    if 'gemini' in str(self.model_name).lower() or str(self.model_name).startswith('google/'):
        logger.info(f"Using special handling for OpenRouter/Gemini model: {self.model_name}")
        # Use raw mode for OpenRouter/Gemini
        output = self.llm.invoke(input_messages)
        output.content = self._remove_think_tags(str(output.content))
        try:
            parsed_json = extract_json_from_model_output(output.content)
            parsed = self.AgentOutput(**parsed_json)
        except (ValueError, ValidationError) as e:
            logger.warning(f'Failed to parse model output: {output} {str(e)}')
            raise ValueError('Could not parse response.')
    elif self.tool_calling_method == 'raw':
        # Existing code for raw mode...
    elif self.tool_calling_method is None:
        # Existing code for tool_calling_method None...
    else:
        # Existing code for other methods...

Environment Configuration

Create or modify your .env file to include your OpenRouter API key:

# Configuration for the OpenRouter API
# Replace the value below with your OpenRouter API key
# You can get an API key at https://openrouter.ai/keys
OPENROUTER_API_KEY=your_openrouter_api_key_here

Then, in your main script, make sure to load this key:

from dotenv import load_dotenv
load_dotenv()

# Get the OpenRouter API key from .env file
openrouter_api_key = os.getenv("OPENROUTER_API_KEY")
if not openrouter_api_key:
    raise ValueError("OPENROUTER_API_KEY not found in environment variables. Please add it to your .env file.")

Known Limitations

1. Some models via OpenRouter may not support all browser-use features, particularly complex function calls.
2. Performance may vary depending on the chosen model.
3. Some errors may occur when processing model responses, requiring additional error handling.

[mtworth]

It would be great to see this OOTB since OpenRouter offers free models + quick ability to test different models.

[MagMueller]

https://github.com/browser-use/browser-use/blob/main/examples/models/openrouter.py