alihan/browser-use-web-ui
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge branch 'main' into feature/arm64-support

Browse Source
This commit is contained in:
Sheldon Aristide
2025-02-05 12:36:41 -05:00
committed by GitHub
parent 1d26f29894 0cdcfd2898
commit 3f0e081637
26 changed files with 859 additions and 1012 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
.env.example
View File

@@ -2,6 +2,7 @@ OPENAI_ENDPOINT=https://api.openai.com/v1
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
ANTHROPIC_ENDPOINT=https://api.anthropic.com
GOOGLE_API_KEY=
@@ -11,6 +12,11 @@ AZURE_OPENAI_API_KEY=
DEEPSEEK_ENDPOINT=https://api.deepseek.com
DEEPSEEK_API_KEY=
MISTRAL_API_KEY=
MISTRAL_ENDPOINT=https://api.mistral.ai/v1
OLLAMA_ENDPOINT=http://localhost:11434
# Set to false to disable anonymized telemetry
ANONYMIZED_TELEMETRY=true
@@ -22,12 +28,16 @@ CHROME_PATH=
CHROME_USER_DATA=
CHROME_DEBUGGING_PORT=9222
CHROME_DEBUGGING_HOST=localhost
CHROME_PERSISTENT_SESSION=false # Set to true to keep browser open between AI tasks
# Set to true to keep browser open between AI tasks
CHROME_PERSISTENT_SESSION=false
# Display settings
RESOLUTION=1920x1080x24 # Format: WIDTHxHEIGHTxDEPTH
RESOLUTION_WIDTH=1920 # Width in pixels
RESOLUTION_HEIGHT=1080 # Height in pixels
# Format: WIDTHxHEIGHTxDEPTH
RESOLUTION=1920x1080x24
# Width in pixels
RESOLUTION_WIDTH=1920
# Height in pixels
RESOLUTION_HEIGHT=1080
# VNC settings
VNC_PASSWORD=youvncpassword

1
Dockerfile
View File

@@ -3,6 +3,7 @@ FROM python:3.11-slim
# Install system dependencies
RUN apt-get update && apt-get install -y \
wget \
netcat-traditional \
gnupg \
curl \
unzip \

107
README.md
View File

@@ -11,7 +11,7 @@ This project builds upon the foundation of the [browser-use](https://github.com/
We would like to officially thank [WarmShao](https://github.com/warmshao) for his contribution to this project.
**WebUI:** is built on Gradio and supports a most of `browser-use` functionalities. This UI is designed to be user-friendly and enables easy interaction with the browser agent.
**WebUI:** is built on Gradio and supports most of `browser-use` functionalities. This UI is designed to be user-friendly and enables easy interaction with the browser agent.
**Expanded LLM Support:** We've integrated support for various Large Language Models (LLMs), including: Gemini, OpenAI, Azure OpenAI, Anthropic, DeepSeek, Ollama etc. And we plan to add support for even more models in the future.
@@ -21,64 +21,93 @@ We would like to officially thank [WarmShao](https://github.com/warmshao) for hi
<video src="https://github.com/user-attachments/assets/56bc7080-f2e3-4367-af22-6bf2245ff6cb" controls="controls">Your browser does not support playing this video!</video>
## Installation Options
## Installation Guide
### Prerequisites
- Python 3.11 or higher
- Git (for cloning the repository)
### Option 1: Local Installation
Read the [quickstart guide](https://docs.browser-use.com/quickstart#prepare-the-environment) or follow the steps below to get started.
> Python 3.11 or higher is required.
#### Step 1: Clone the Repository
```bash
git clone https://github.com/browser-use/web-ui.git
cd web-ui
```
First, we recommend using [uv](https://docs.astral.sh/uv/) to setup the Python environment.
#### Step 2: Set Up Python Environment
We recommend using [uv](https://docs.astral.sh/uv/) for managing the Python environment.
Using uv (recommended):
```bash
uv venv --python 3.11
```
and activate it with:
Activate the virtual environment:
- Windows (Command Prompt):
```cmd
.venv\Scripts\activate
```
- Windows (PowerShell):
```powershell
.\.venv\Scripts\Activate.ps1
```
- macOS/Linux:
```bash
source .venv/bin/activate
```
Install the dependencies:
#### Step 3: Install Dependencies
Install Python packages:
```bash
uv pip install -r requirements.txt
```
Then install playwright:
Install Playwright:
```bash
playwright install
```
#### Step 4: Configure Environment
1. Create a copy of the example environment file:
- Windows (Command Prompt):
```bash
copy .env.example .env
```
- macOS/Linux/Windows (PowerShell):
```bash
cp .env.example .env
```
2. Open `.env` in your preferred text editor and add your API keys and other settings
### Option 2: Docker Installation
1. **Prerequisites:**
- Docker and Docker Compose installed on your system
- Git to clone the repository
#### Prerequisites
- Docker and Docker Compose installed
- [Docker Desktop](https://www.docker.com/products/docker-desktop/) (For Windows/macOS)
- [Docker Engine](https://docs.docker.com/engine/install/) and [Docker Compose](https://docs.docker.com/compose/install/) (For Linux)
2. **Setup:**
#### Installation Steps
1. Clone the repository:
```bash
# Clone the repository
git clone https://github.com/browser-use/web-ui.git
cd web-ui
# Copy and configure environment variables
cp .env.example .env
# Edit .env with your preferred text editor and add your API keys
```
3. **Run with Docker:**
2. Create and configure environment file:
- Windows (Command Prompt):
```bash
# Build and start the container with default settings (browser closes after AI tasks)
docker compose up --build
# Or run with persistent browser (browser stays open between AI tasks)
CHROME_PERSISTENT_SESSION=true docker compose up --build
copy .env.example .env
```
- macOS/Linux/Windows (PowerShell):
```bash
cp .env.example .env
```
Edit `.env` with your preferred text editor and add your API keys
feature/arm64-support
4. **Access the Application:**
- WebUI: `http://localhost:7788`
- VNC Viewer (to see browser interactions): `http://localhost:6080/vnc.html`
@@ -86,16 +115,32 @@ playwright install
Default VNC password is "vncpassword". You can change it by setting the `VNC_PASSWORD` environment variable in your `.env` file.
3. Run with Docker:
```bash
# Build and start the container with default settings (browser closes after AI tasks)
docker compose up --build
```
```bash
# Or run with persistent browser (browser stays open between AI tasks)
CHROME_PERSISTENT_SESSION=true docker compose up --build
```
4. Access the Application:
- Web Interface: Open `http://localhost:7788` in your browser
- VNC Viewer (for watching browser interactions): Open `http://localhost:6080/vnc.html`
- Default VNC password: "youvncpassword"
- Can be changed by setting `VNC_PASSWORD` in your `.env` file
## Usage
### Local Setup
1. Copy `.env.example` to `.env` and set your environment variables, including API keys for the LLM. `cp .env.example .env`
2. **Run the WebUI:**
1. **Run the WebUI:**
After completing the installation steps above, start the application:
```bash
python webui.py --ip 127.0.0.1 --port 7788
```
4. WebUI options:
2. WebUI options:
- `--ip`: The IP address to bind the WebUI to. Default is `127.0.0.1`.
- `--port`: The port to bind the WebUI to. Default is `7788`.
- `--theme`: The theme for the user interface. Default is `Ocean`.
@@ -109,7 +154,7 @@ playwright install
- `--dark-mode`: Enables dark mode for the user interface.
3. **Access the WebUI:** Open your web browser and navigate to `http://127.0.0.1:7788`.
4. **Using Your Own Browser(Optional):**
- Set `CHROME_PATH` to the executable path of your browser and `CHROME_USER_DATA` to the user data directory of your browser.
- Set `CHROME_PATH` to the executable path of your browser and `CHROME_USER_DATA` to the user data directory of your browser. Leave `CHROME_USER_DATA` empty if you want to use local user data.
- Windows
```env
CHROME_PATH="C:\Program Files\Google\Chrome\Application\chrome.exe"
@@ -119,7 +164,7 @@ playwright install
- Mac
```env
CHROME_PATH="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
CHROME_USER_DATA="~/Library/Application Support/Google/Chrome/Profile 1"
CHROME_USER_DATA="/Users/YourUsername/Library/Application Support/Google/Chrome"
```
- Close all Chrome windows
- Open the WebUI in a non-Chrome browser, such as Firefox or Edge. This is important because the persistent browser context will use the Chrome data when running the agent.
@@ -185,6 +230,6 @@ playwright install
```
## Changelog
- [x] **2025/01/26:** Thanks to @vvincent1234. Now browser-use-webui can combine with DeepSeek-r1 to engage in deep thinking!
- [x] **2025/01/10:** Thanks to @casistack. Now we have Docker Setup option and also Support keep browser open between tasks.[Video tutorial demo](https://github.com/browser-use/web-ui/issues/1#issuecomment-2582511750).
- [x] **2025/01/06:** Thanks to @richard-devbot. A New and Well-Designed WebUI is released. [Video tutorial demo](https://github.com/warmshao/browser-use-webui/issues/1#issuecomment-2573393113).

19
SECURITY.md Normal file
View File

@@ -0,0 +1,19 @@
## Reporting Security Issues
If you believe you have found a security vulnerability in browser-use, please report it through coordinated disclosure.
**Please do not report security vulnerabilities through the repository issues, discussions, or pull requests.**
Instead, please open a new [Github security advisory](https://github.com/browser-use/web-ui/security/advisories/new).
Please include as much of the information listed below as you can to help me better understand and resolve the issue:
* The type of issue (e.g., buffer overflow, SQL injection, or cross-site scripting)
* Full paths of source file(s) related to the manifestation of the issue
* The location of the affected source code (tag/branch/commit or direct URL)
* Any special configuration required to reproduce the issue
* Step-by-step instructions to reproduce the issue
* Proof-of-concept or exploit code (if possible)
* Impact of the issue, including how an attacker might exploit the issue
This information will help me triage your report more quickly.

1
docker-compose.yml
View File

@@ -1,5 +1,6 @@
services:
browser-use-webui:
platform: linux/amd64
build:
context: .
dockerfile: ${DOCKERFILE:-Dockerfile}

9
requirements.txt
View File

@@ -1,6 +1,5 @@
browser-use==0.1.19
langchain-google-genai==2.0.8
browser-use==0.1.29
pyperclip==1.9.0
gradio==5.9.1
langchain-ollama==0.2.2
langchain-openai==0.2.14
gradio==5.10.0
json-repair
langchain-mistralai==0.2.4

6
src/__init__.py
View File

@@ -1,6 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: __init__.py.py

6
src/agent/__init__.py
View File

@@ -1,6 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: __init__.py.py

432
src/agent/custom_agent.py
View File

@@ -1,23 +1,18 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: custom_agent.py
import json
import logging
import pdb
import traceback
from typing import Optional, Type
from typing import Optional, Type, List, Dict, Any, Callable
from PIL import Image, ImageDraw, ImageFont
import os
import base64
import io
from browser_use.agent.prompts import SystemPrompt
import platform
from browser_use.agent.prompts import SystemPrompt, AgentMessagePrompt
from browser_use.agent.service import Agent
from browser_use.agent.views import (
ActionResult,
ActionModel,
AgentHistoryList,
AgentOutput,
AgentHistory,
@@ -29,13 +24,14 @@ from browser_use.controller.service import Controller
from browser_use.telemetry.views import (
AgentEndTelemetryEvent,
AgentRunTelemetryEvent,
AgentStepErrorTelemetryEvent,
AgentStepTelemetryEvent,
)
from browser_use.utils import time_execution_async
from langchain_core.language_models.chat_models import BaseChatModel
from langchain_core.messages import (
BaseMessage,
)
from json_repair import repair_json
from src.utils.agent_state import AgentState
from .custom_massage_manager import CustomMassageManager
@@ -58,6 +54,7 @@ class CustomAgent(Agent):
max_failures: int = 5,
retry_delay: int = 10,
system_prompt_class: Type[SystemPrompt] = SystemPrompt,
agent_prompt_class: Type[AgentMessagePrompt] = AgentMessagePrompt,
max_input_tokens: int = 128000,
validate_output: bool = False,
include_attributes: list[str] = [
@@ -76,6 +73,11 @@ class CustomAgent(Agent):
max_actions_per_step: int = 10,
tool_call_in_content: bool = True,
agent_state: AgentState = None,
initial_actions: Optional[List[Dict[str, Dict[str, Any]]]] = None,
# Cloud Callbacks
register_new_step_callback: Callable[['BrowserState', 'AgentOutput', int], None] | None = None,
register_done_callback: Callable[['AgentHistoryList'], None] | None = None,
tool_calling_method: Optional[str] = 'auto',
):
super().__init__(
task=task,
@@ -94,26 +96,36 @@ class CustomAgent(Agent):
max_error_length=max_error_length,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content,
initial_actions=initial_actions,
register_new_step_callback=register_new_step_callback,
register_done_callback=register_done_callback,
tool_calling_method=tool_calling_method
)
if self.llm.model_name in ["deepseek-reasoner"]:
self.use_function_calling = False
# TODO: deepseek-reasoner only support 64000 context
if self.model_name in ["deepseek-reasoner"] or "deepseek-r1" in self.model_name:
# deepseek-reasoner does not support function calling
self.use_deepseek_r1 = True
# deepseek-reasoner only support 64000 context
self.max_input_tokens = 64000
else:
self.use_function_calling = True
self.use_deepseek_r1 = False
# record last actions
self._last_actions = None
# custom new info
self.add_infos = add_infos
# agent_state for Stop
self.agent_state = agent_state
self.agent_prompt_class = agent_prompt_class
self.message_manager = CustomMassageManager(
llm=self.llm,
task=self.task,
action_descriptions=self.controller.registry.get_prompt_description(),
system_prompt_class=self.system_prompt_class,
agent_prompt_class=agent_prompt_class,
max_input_tokens=self.max_input_tokens,
include_attributes=self.include_attributes,
max_error_length=self.max_error_length,
max_actions_per_step=self.max_actions_per_step,
tool_call_in_content=tool_call_in_content,
use_function_calling=self.use_function_calling
max_actions_per_step=self.max_actions_per_step
)
def _setup_action_models(self) -> None:
@@ -172,52 +184,35 @@ class CustomAgent(Agent):
@time_execution_async("--get_next_action")
async def get_next_action(self, input_messages: list[BaseMessage]) -> AgentOutput:
"""Get next action from LLM based on current state"""
if self.use_function_calling:
try:
structured_llm = self.llm.with_structured_output(self.AgentOutput, include_raw=True)
response: dict[str, Any] = await structured_llm.ainvoke(input_messages) # type: ignore
messages_to_process = (
self.message_manager.merge_successive_human_messages(input_messages)
if self.use_deepseek_r1
else input_messages
)
parsed: AgentOutput = response['parsed']
# cut the number of actions to max_actions_per_step
parsed.action = parsed.action[: self.max_actions_per_step]
self._log_response(parsed)
self.n_steps += 1
ai_message = self.llm.invoke(messages_to_process)
self.message_manager._add_message_with_tokens(ai_message)
return parsed
except Exception as e:
# If something goes wrong, try to invoke the LLM again without structured output,
# and Manually parse the response. Temporarily solution for DeepSeek
ret = self.llm.invoke(input_messages)
if isinstance(ret.content, list):
parsed_json = json.loads(ret.content[0].replace("```json", "").replace("```", ""))
if self.use_deepseek_r1:
logger.info("🤯 Start Deep Thinking: ")
logger.info(ai_message.reasoning_content)
logger.info("🤯 End Deep Thinking")
if isinstance(ai_message.content, list):
ai_content = ai_message.content[0]
else:
parsed_json = json.loads(ret.content.replace("```json", "").replace("```", ""))
ai_content = ai_message.content
ai_content = ai_content.replace("```json", "").replace("```", "")
ai_content = repair_json(ai_content)
parsed_json = json.loads(ai_content)
parsed: AgentOutput = self.AgentOutput(**parsed_json)
if parsed is None:
raise ValueError(f'Could not parse response.')
logger.debug(ai_message.content)
raise ValueError('Could not parse response.')
# cut the number of actions to max_actions_per_step
parsed.action = parsed.action[: self.max_actions_per_step]
self._log_response(parsed)
self.n_steps += 1
return parsed
else:
ret = self.llm.invoke(input_messages)
if not self.use_function_calling:
self.message_manager._add_message_with_tokens(ret)
logger.info(f"🤯 Start Deep Thinking: ")
logger.info(ret.reasoning_content)
logger.info(f"🤯 End Deep Thinking")
if isinstance(ret.content, list):
parsed_json = json.loads(ret.content[0].replace("```json", "").replace("```", ""))
else:
parsed_json = json.loads(ret.content.replace("```json", "").replace("```", ""))
parsed: AgentOutput = self.AgentOutput(**parsed_json)
if parsed is None:
raise ValueError(f'Could not parse response.')
# cut the number of actions to max_actions_per_step
# Limit actions to maximum allowed per step
parsed.action = parsed.action[: self.max_actions_per_step]
self._log_response(parsed)
self.n_steps += 1
@@ -234,50 +229,200 @@ class CustomAgent(Agent):
try:
state = await self.browser_context.get_state(use_vision=self.use_vision)
self.message_manager.add_state_message(state, self._last_result, step_info)
self.message_manager.add_state_message(state, self._last_actions, self._last_result, step_info)
input_messages = self.message_manager.get_messages()
try:
model_output = await self.get_next_action(input_messages)
if self.register_new_step_callback:
self.register_new_step_callback(state, model_output, self.n_steps)
self.update_step_info(model_output, step_info)
logger.info(f"🧠 All Memory: \n{step_info.memory}")
self._save_conversation(input_messages, model_output)
if self.use_function_calling:
self.message_manager._remove_last_state_message() # we dont want the whole state in the chat history
self.message_manager.add_model_output(model_output)
if self.model_name != "deepseek-reasoner":
# remove prev message
self.message_manager._remove_state_message_by_index(-1)
except Exception as e:
# model call failed, remove last state message from history
self.message_manager._remove_state_message_by_index(-1)
raise e
actions: list[ActionModel] = model_output.action
result: list[ActionResult] = await self.controller.multi_act(
model_output.action, self.browser_context
actions, self.browser_context
)
if len(result) != len(model_output.action):
for ri in range(len(result), len(model_output.action)):
if len(result) != len(actions):
# I think something changes, such information should let LLM know
for ri in range(len(result), len(actions)):
result.append(ActionResult(extracted_content=None,
include_in_memory=True,
error=f"{model_output.action[ri].model_dump_json(exclude_unset=True)} is Failed to execute. \
Something new appeared after action {model_output.action[len(result) - 1].model_dump_json(exclude_unset=True)}",
error=f"{actions[ri].model_dump_json(exclude_unset=True)} is Failed to execute. \
Something new appeared after action {actions[len(result) - 1].model_dump_json(exclude_unset=True)}",
is_done=False))
if len(actions) == 0:
# TODO: fix no action case
result = [ActionResult(is_done=True, extracted_content=step_info.memory, include_in_memory=True)]
self._last_result = result
self._last_actions = actions
if len(result) > 0 and result[-1].is_done:
logger.info(f"📄 Result: {result[-1].extracted_content}")
self.consecutive_failures = 0
except Exception as e:
result = self._handle_step_error(e)
result = await self._handle_step_error(e)
self._last_result = result
finally:
actions = [a.model_dump(exclude_unset=True) for a in model_output.action] if model_output else []
self.telemetry.capture(
AgentStepTelemetryEvent(
agent_id=self.agent_id,
step=self.n_steps,
actions=actions,
consecutive_failures=self.consecutive_failures,
step_error=[r.error for r in result if r.error] if result else ['No result'],
)
)
if not result:
return
for r in result:
if r.error:
self.telemetry.capture(
AgentStepErrorTelemetryEvent(
agent_id=self.agent_id,
error=r.error,
)
)
if state:
self._make_history_item(model_output, state, result)
async def run(self, max_steps: int = 100) -> AgentHistoryList:
"""Execute the task with maximum number of steps"""
try:
self._log_agent_run()
# Execute initial actions if provided
if self.initial_actions:
result = await self.controller.multi_act(self.initial_actions, self.browser_context, check_for_new_elements=False)
self._last_result = result
step_info = CustomAgentStepInfo(
task=self.task,
add_infos=self.add_infos,
step_number=1,
max_steps=max_steps,
memory="",
task_progress="",
future_plans=""
)
for step in range(max_steps):
# 1) Check if stop requested
if self.agent_state and self.agent_state.is_stop_requested():
logger.info("🛑 Stop requested by user")
self._create_stop_history_item()
break
# 2) Store last valid state before step
if self.browser_context and self.agent_state:
state = await self.browser_context.get_state(use_vision=self.use_vision)
self.agent_state.set_last_valid_state(state)
if self._too_many_failures():
break
# 3) Do the step
await self.step(step_info)
if self.history.is_done():
if (
self.validate_output and step < max_steps - 1
): # if last step, we dont need to validate
if not await self._validate_output():
continue
logger.info("✅ Task completed successfully")
break
else:
logger.info("❌ Failed to complete task in maximum steps")
return self.history
finally:
self.telemetry.capture(
AgentEndTelemetryEvent(
agent_id=self.agent_id,
success=self.history.is_done(),
steps=self.n_steps,
max_steps_reached=self.n_steps >= max_steps,
errors=self.history.errors(),
)
)
if not self.injected_browser_context:
await self.browser_context.close()
if not self.injected_browser and self.browser:
await self.browser.close()
if self.generate_gif:
output_path: str = 'agent_history.gif'
if isinstance(self.generate_gif, str):
output_path = self.generate_gif
self.create_history_gif(output_path=output_path)
def _create_stop_history_item(self):
"""Create a history item for when the agent is stopped."""
try:
# Attempt to retrieve the last valid state from agent_state
state = None
if self.agent_state:
last_state = self.agent_state.get_last_valid_state()
if last_state:
# Convert to BrowserStateHistory
state = BrowserStateHistory(
url=getattr(last_state, 'url', ""),
title=getattr(last_state, 'title', ""),
tabs=getattr(last_state, 'tabs', []),
interacted_element=[None],
screenshot=getattr(last_state, 'screenshot', None)
)
else:
state = self._create_empty_state()
else:
state = self._create_empty_state()
# Create a final item in the agent history indicating done
stop_history = AgentHistory(
model_output=None,
state=state,
result=[ActionResult(extracted_content=None, error=None, is_done=True)]
)
self.history.history.append(stop_history)
except Exception as e:
logger.error(f"Error creating stop history item: {e}")
# Create empty state as fallback
state = self._create_empty_state()
stop_history = AgentHistory(
model_output=None,
state=state,
result=[ActionResult(extracted_content=None, error=None, is_done=True)]
)
self.history.history.append(stop_history)
def _convert_to_browser_state_history(self, browser_state):
return BrowserStateHistory(
url=getattr(browser_state, 'url', ""),
title=getattr(browser_state, 'title', ""),
tabs=getattr(browser_state, 'tabs', []),
interacted_element=[None],
screenshot=getattr(browser_state, 'screenshot', None)
)
def _create_empty_state(self):
return BrowserStateHistory(
url="",
title="",
tabs=[],
interacted_element=[None],
screenshot=None
)
def create_history_gif(
self,
output_path: str = 'agent_history.gif',
@@ -310,10 +455,9 @@ class CustomAgent(Agent):
for font_name in font_options:
try:
import platform
if platform.system() == "Windows":
if platform.system() == 'Windows':
# Need to specify the abs font path on Windows
font_name = os.path.join(os.getenv("WIN_FONT_DIR", "C:\\Windows\\Fonts"), font_name + ".ttf")
font_name = os.path.join(os.getenv('WIN_FONT_DIR', 'C:\\Windows\\Fonts'), font_name + '.ttf')
regular_font = ImageFont.truetype(font_name, font_size)
title_font = ImageFont.truetype(font_name, title_font_size)
goal_font = ImageFont.truetype(font_name, goal_font_size)
@@ -391,133 +535,3 @@ class CustomAgent(Agent):
logger.info(f'Created GIF at {output_path}')
else:
logger.warning('No images found in history to create GIF')
async def run(self, max_steps: int = 100) -> AgentHistoryList:
"""Execute the task with maximum number of steps"""
try:
logger.info(f"🚀 Starting task: {self.task}")
self.telemetry.capture(
AgentRunTelemetryEvent(
agent_id=self.agent_id,
task=self.task,
)
)
step_info = CustomAgentStepInfo(
task=self.task,
add_infos=self.add_infos,
step_number=1,
max_steps=max_steps,
memory="",
task_progress="",
future_plans=""
)
for step in range(max_steps):
# 1) Check if stop requested
if self.agent_state and self.agent_state.is_stop_requested():
logger.info("🛑 Stop requested by user")
self._create_stop_history_item()
break
# 2) Store last valid state before step
if self.browser_context and self.agent_state:
state = await self.browser_context.get_state(use_vision=self.use_vision)
self.agent_state.set_last_valid_state(state)
if self._too_many_failures():
break
# 3) Do the step
await self.step(step_info)
if self.history.is_done():
if (
self.validate_output and step < max_steps - 1
): # if last step, we dont need to validate
if not await self._validate_output():
continue
logger.info("✅ Task completed successfully")
break
else:
logger.info("❌ Failed to complete task in maximum steps")
return self.history
finally:
self.telemetry.capture(
AgentEndTelemetryEvent(
agent_id=self.agent_id,
task=self.task,
success=self.history.is_done(),
steps=len(self.history.history),
)
)
if not self.injected_browser_context:
await self.browser_context.close()
if not self.injected_browser and self.browser:
await self.browser.close()
if self.generate_gif:
self.create_history_gif()
def _create_stop_history_item(self):
"""Create a history item for when the agent is stopped."""
try:
# Attempt to retrieve the last valid state from agent_state
state = None
if self.agent_state:
last_state = self.agent_state.get_last_valid_state()
if last_state:
# Convert to BrowserStateHistory
state = BrowserStateHistory(
url=getattr(last_state, 'url', ""),
title=getattr(last_state, 'title', ""),
tabs=getattr(last_state, 'tabs', []),
interacted_element=[None],
screenshot=getattr(last_state, 'screenshot', None)
)
else:
state = self._create_empty_state()
else:
state = self._create_empty_state()
# Create a final item in the agent history indicating done
stop_history = AgentHistory(
model_output=None,
state=state,
result=[ActionResult(extracted_content=None, error=None, is_done=True)]
)
self.history.history.append(stop_history)
except Exception as e:
logger.error(f"Error creating stop history item: {e}")
# Create empty state as fallback
state = self._create_empty_state()
stop_history = AgentHistory(
model_output=None,
state=state,
result=[ActionResult(extracted_content=None, error=None, is_done=True)]
)
self.history.history.append(stop_history)
def _convert_to_browser_state_history(self, browser_state):
return BrowserStateHistory(
url=getattr(browser_state, 'url', ""),
title=getattr(browser_state, 'title', ""),
tabs=getattr(browser_state, 'tabs', []),
interacted_element=[None],
screenshot=getattr(browser_state, 'screenshot', None)
)
def _create_empty_state(self):
return BrowserStateHistory(
url="",
title="",
tabs=[],
interacted_element=[None],
screenshot=None
)

103
src/agent/custom_massage_manager.py
View File

@@ -1,9 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: custom_massage_manager.py
from __future__ import annotations
import logging
@@ -11,15 +5,20 @@ from typing import List, Optional, Type
from browser_use.agent.message_manager.service import MessageManager
from browser_use.agent.message_manager.views import MessageHistory
from browser_use.agent.prompts import SystemPrompt
from browser_use.agent.views import ActionResult, AgentStepInfo
from browser_use.agent.prompts import SystemPrompt, AgentMessagePrompt
from browser_use.agent.views import ActionResult, AgentStepInfo, ActionModel
from browser_use.browser.views import BrowserState
from langchain_core.language_models import BaseChatModel
from langchain_anthropic import ChatAnthropic
from langchain_core.language_models import BaseChatModel
from langchain_core.messages import (
AIMessage,
BaseMessage,
HumanMessage,
AIMessage
ToolMessage
)
from langchain_openai import ChatOpenAI
from ..utils.llm import DeepSeekR1ChatOpenAI
from .custom_prompts import CustomAgentMessagePrompt
logger = logging.getLogger(__name__)
@@ -32,14 +31,14 @@ class CustomMassageManager(MessageManager):
task: str,
action_descriptions: str,
system_prompt_class: Type[SystemPrompt],
agent_prompt_class: Type[AgentMessagePrompt],
max_input_tokens: int = 128000,
estimated_tokens_per_character: int = 3,
estimated_characters_per_token: int = 3,
image_tokens: int = 800,
include_attributes: list[str] = [],
max_error_length: int = 400,
max_actions_per_step: int = 10,
tool_call_in_content: bool = False,
use_function_calling: bool = True
message_context: Optional[str] = None
):
super().__init__(
llm=llm,
@@ -47,72 +46,72 @@ class CustomMassageManager(MessageManager):
action_descriptions=action_descriptions,
system_prompt_class=system_prompt_class,
max_input_tokens=max_input_tokens,
estimated_tokens_per_character=estimated_tokens_per_character,
estimated_characters_per_token=estimated_characters_per_token,
image_tokens=image_tokens,
include_attributes=include_attributes,
max_error_length=max_error_length,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content,
message_context=message_context
)
self.use_function_calling = use_function_calling
self.agent_prompt_class = agent_prompt_class
# Custom: Move Task info to state_message
self.history = MessageHistory()
self._add_message_with_tokens(self.system_prompt)
if self.use_function_calling:
tool_calls = [
{
'name': 'CustomAgentOutput',
'args': {
'current_state': {
'prev_action_evaluation': 'Unknown - No previous actions to evaluate.',
'important_contents': '',
'completed_contents': '',
'thought': 'Now Google is open. Need to type OpenAI to search.',
'summary': 'Type OpenAI to search.',
},
'action': [],
},
'id': '',
'type': 'tool_call',
}
]
if self.tool_call_in_content:
# openai throws error if tool_calls are not responded -> move to content
example_tool_call = AIMessage(
content=f'{tool_calls}',
tool_calls=[],
)
else:
example_tool_call = AIMessage(
content=f'',
tool_calls=tool_calls,
)
self._add_message_with_tokens(example_tool_call)
if self.message_context:
context_message = HumanMessage(content=self.message_context)
self._add_message_with_tokens(context_message)
def cut_messages(self):
"""Get current message list, potentially trimmed to max tokens"""
diff = self.history.total_tokens - self.max_input_tokens
i = 1 # start from 1 to keep system message in history
while diff > 0 and i < len(self.history.messages):
self.history.remove_message(i)
min_message_len = 2 if self.message_context is not None else 1
while diff > 0 and len(self.history.messages) > min_message_len:
self.history.remove_message(min_message_len) # alway remove the oldest message
diff = self.history.total_tokens - self.max_input_tokens
i += 1
def add_state_message(
self,
state: BrowserState,
actions: Optional[List[ActionModel]] = None,
result: Optional[List[ActionResult]] = None,
step_info: Optional[AgentStepInfo] = None,
) -> None:
"""Add browser state as human message"""
# otherwise add state message and result to next message (which will not stay in memory)
state_message = CustomAgentMessagePrompt(
state_message = self.agent_prompt_class(
state,
actions,
result,
include_attributes=self.include_attributes,
max_error_length=self.max_error_length,
step_info=step_info,
).get_user_message()
self._add_message_with_tokens(state_message)
def _count_text_tokens(self, text: str) -> int:
if isinstance(self.llm, (ChatOpenAI, ChatAnthropic, DeepSeekR1ChatOpenAI)):
try:
tokens = self.llm.get_num_tokens(text)
except Exception:
tokens = (
len(text) // self.estimated_characters_per_token
) # Rough estimate if no tokenizer available
else:
tokens = (
len(text) // self.estimated_characters_per_token
) # Rough estimate if no tokenizer available
return tokens
def _remove_state_message_by_index(self, remove_ind=-1) -> None:
"""Remove last state message from history"""
i = len(self.history.messages) - 1
remove_cnt = 0
while i >= 0:
if isinstance(self.history.messages[i].message, HumanMessage):
remove_cnt += 1
if remove_cnt == abs(remove_ind):
self.history.remove_message(i)
break
i -= 1

75
src/agent/custom_prompts.py
View File

@@ -1,13 +1,8 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: custom_prompts.py
import pdb
from typing import List, Optional
from browser_use.agent.prompts import SystemPrompt
from browser_use.agent.views import ActionResult
from browser_use.agent.prompts import SystemPrompt, AgentMessagePrompt
from browser_use.agent.views import ActionResult, ActionModel
from browser_use.browser.views import BrowserState
from langchain_core.messages import HumanMessage, SystemMessage
@@ -19,24 +14,19 @@ class CustomSystemPrompt(SystemPrompt):
"""
Returns the important rules for the agent.
"""
text = """
text = r"""
1. RESPONSE FORMAT: You must ALWAYS respond with valid JSON in this exact format:
{
"current_state": {
"prev_action_evaluation": "Success|Failed|Unknown - Analyze the current elements and the image to check if the previous goals/actions are successful like intended by the task. Ignore the action result. The website is the ground truth. Also mention if something unexpected happened like new suggestions in an input field. Shortly state why/why not. Note that the result you output must be consistent with the reasoning you output afterwards. If you consider it to be 'Failed,' you should reflect on this during your thought.",
"important_contents": "Output important contents closely related to user\'s instruction or task on the current page. If there is, please output the contents. If not, please output empty string ''.",
"important_contents": "Output important contents closely related to user\'s instruction on the current page. If there is, please output the contents. If not, please output empty string ''.",
"task_progress": "Task Progress is a general summary of the current contents that have been completed. Just summarize the contents that have been actually completed based on the content at current step and the history operations. Please list each completed item individually, such as: 1. Input username. 2. Input Password. 3. Click confirm button. Please return string type not a list.",
"future_plans": "Based on the user's request and the current state, outline the remaining steps needed to complete the task. This should be a concise list of actions yet to be performed, such as: 1. Select a date. 2. Choose a specific time slot. 3. Confirm booking. Please return string type not a list.",
"thought": "Think about the requirements that have been completed in previous operations and the requirements that need to be completed in the next one operation. If your output of prev_action_evaluation is 'Failed', please reflect and output your reflection here.",
"summary": "Please generate a brief natural language description for the operation in next actions based on your Thought."
},
"action": [
{
"action_name": {
// action-specific parameters
}
},
// ... more actions in sequence
* actions in sequences, please refer to **Common action sequences**. Each output action MUST be formated as: \{action_name\: action_params\}*
]
}
@@ -49,7 +39,6 @@ class CustomSystemPrompt(SystemPrompt):
{"click_element": {"index": 3}}
]
- Navigation and extraction: [
{"open_new_tab": {}},
{"go_to_url": {"url": "https://example.com"}},
{"extract_page_content": {}}
]
@@ -67,7 +56,7 @@ class CustomSystemPrompt(SystemPrompt):
- Use scroll to find elements you are looking for
5. TASK COMPLETION:
- If you think all the requirements of user\'s instruction have been completed and no further operation is required, output the done action to terminate the operation process.
- If you think all the requirements of user\'s instruction have been completed and no further operation is required, output the **Done** action to terminate the operation process.
- Don't hallucinate actions.
- If the task requires specific information - make sure to include everything in the done function. This is what the user will see.
- If you are running out of steps (current step), think about speeding it up, and ALWAYS use the done action as the last action.
@@ -132,7 +121,7 @@ class CustomSystemPrompt(SystemPrompt):
AGENT_PROMPT = f"""You are a precise browser automation agent that interacts with websites through structured commands. Your role is to:
1. Analyze the provided webpage elements and structure
2. Plan a sequence of actions to accomplish the given task
3. Respond with valid JSON containing your action sequence and state assessment
3. Your final result MUST be a valid JSON as the **RESPONSE FORMAT** described, containing your action sequence and state assessment, No need extra content to expalin.
Current date and time: {time_str}
@@ -147,33 +136,54 @@ class CustomSystemPrompt(SystemPrompt):
return SystemMessage(content=AGENT_PROMPT)
class CustomAgentMessagePrompt:
class CustomAgentMessagePrompt(AgentMessagePrompt):
def __init__(
self,
state: BrowserState,
actions: Optional[List[ActionModel]] = None,
result: Optional[List[ActionResult]] = None,
include_attributes: list[str] = [],
max_error_length: int = 400,
step_info: Optional[CustomAgentStepInfo] = None,
):
self.state = state
self.result = result
self.max_error_length = max_error_length
self.include_attributes = include_attributes
self.step_info = step_info
super(CustomAgentMessagePrompt, self).__init__(state=state,
result=result,
include_attributes=include_attributes,
max_error_length=max_error_length,
step_info=step_info
)
self.actions = actions
def get_user_message(self) -> HumanMessage:
if self.step_info:
step_info_description = f'Current step: {self.step_info.step_number + 1}/{self.step_info.max_steps}'
step_info_description = f'Current step: {self.step_info.step_number}/{self.step_info.max_steps}\n'
else:
step_info_description = ''
elements_text = self.state.element_tree.clickable_elements_to_string(include_attributes=self.include_attributes)
if not elements_text:
has_content_above = (self.state.pixels_above or 0) > 0
has_content_below = (self.state.pixels_below or 0) > 0
if elements_text != '':
if has_content_above:
elements_text = (
f'... {self.state.pixels_above} pixels above - scroll or extract content to see more ...\n{elements_text}'
)
else:
elements_text = f'[Start of page]\n{elements_text}'
if has_content_below:
elements_text = (
f'{elements_text}\n... {self.state.pixels_below} pixels below - scroll or extract content to see more ...'
)
else:
elements_text = f'{elements_text}\n[End of page]'
else:
elements_text = 'empty page'
state_description = f"""
{step_info_description}
1. Task: {self.step_info.task}
1. Task: {self.step_info.task}.
2. Hints(Optional):
{self.step_info.add_infos}
3. Memory:
@@ -185,15 +195,20 @@ class CustomAgentMessagePrompt:
{elements_text}
"""
if self.result:
if self.actions and self.result:
state_description += "\n **Previous Actions** \n"
state_description += f'Previous step: {self.step_info.step_number-1}/{self.step_info.max_steps} \n'
for i, result in enumerate(self.result):
action = self.actions[i]
state_description += f"Previous action {i + 1}/{len(self.result)}: {action.model_dump_json(exclude_unset=True)}\n"
if result.include_in_memory:
if result.extracted_content:
state_description += f"\nResult of action {i + 1}/{len(self.result)}: {result.extracted_content}"
state_description += f"Result of previous action {i + 1}/{len(self.result)}: {result.extracted_content}\n"
if result.error:
# only use last 300 characters of error
error = result.error[-self.max_error_length:]
state_description += (
f"\nError of action {i + 1}/{len(self.result)}: ...{error}"
f"Error of previous action {i + 1}/{len(self.result)}: ...{error}\n"
)
if self.state.screenshot:

8
src/agent/custom_views.py
View File

@@ -1,9 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: custom_views.py
from dataclasses import dataclass
from typing import Type
@@ -51,7 +45,7 @@ class CustomAgentOutput(AgentOutput):
) -> Type["CustomAgentOutput"]:
"""Extend actions with custom actions"""
return create_model(
"AgentOutput",
"CustomAgentOutput",
__base__=CustomAgentOutput,
action=(
list[custom_actions],

6
src/browser/__init__.py
View File

@@ -1,6 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: __init__.py.py

30
src/browser/config.py
View File

@@ -1,30 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/6
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: config.py
import os
from dataclasses import dataclass
from typing import Optional
@dataclass
class BrowserPersistenceConfig:
"""Configuration for browser persistence"""
persistent_session: bool = False
user_data_dir: Optional[str] = None
debugging_port: Optional[int] = None
debugging_host: Optional[str] = None
@classmethod
def from_env(cls) -> "BrowserPersistenceConfig":
"""Create config from environment variables"""
return cls(
persistent_session=os.getenv("CHROME_PERSISTENT_SESSION", "").lower()
== "true",
user_data_dir=os.getenv("CHROME_USER_DATA"),
debugging_port=int(os.getenv("CHROME_DEBUGGING_PORT", "9222")),
debugging_host=os.getenv("CHROME_DEBUGGING_HOST", "localhost"),
)

58
src/browser/custom_browser.py
View File

@@ -1,9 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: browser.py
import asyncio
import pdb
@@ -20,7 +14,6 @@ from browser_use.browser.context import BrowserContext, BrowserContextConfig
from playwright.async_api import BrowserContext as PlaywrightBrowserContext
import logging
from .config import BrowserPersistenceConfig
from .custom_context import CustomBrowserContext
logger = logging.getLogger(__name__)
@@ -33,12 +26,10 @@ class CustomBrowser(Browser):
) -> CustomBrowserContext:
return CustomBrowserContext(config=config, browser=self)
async def _setup_browser(self, playwright: Playwright) -> PlaywrightBrowser:
async def _setup_browser_with_instance(self, playwright: Playwright) -> PlaywrightBrowser:
"""Sets up and returns a Playwright Browser instance with anti-detection measures."""
if self.config.wss_url:
browser = await playwright.chromium.connect(self.config.wss_url)
return browser
elif self.config.chrome_instance_path:
if not self.config.chrome_instance_path:
raise ValueError('Chrome instance path is required')
import subprocess
import requests
@@ -61,12 +52,12 @@ class CustomBrowser(Browser):
[
self.config.chrome_instance_path,
'--remote-debugging-port=9222',
],
] + self.config.extra_chromium_args,
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
# Attempt to connect again after starting a new instance
# try to connect first in case the browser have not started
for _ in range(10):
try:
response = requests.get('http://localhost:9222/json/version', timeout=2)
@@ -76,6 +67,7 @@ class CustomBrowser(Browser):
pass
await asyncio.sleep(1)
# Attempt to connect again after starting a new instance
try:
browser = await playwright.chromium.connect_over_cdp(
endpoint_url='http://localhost:9222',
@@ -87,41 +79,3 @@ class CustomBrowser(Browser):
raise RuntimeError(
' To start chrome in Debug mode, you need to close all existing Chrome instances and try again otherwise we can not connect to the instance.'
)
else:
try:
disable_security_args = []
if self.config.disable_security:
disable_security_args = [
'--disable-web-security',
'--disable-site-isolation-trials',
'--disable-features=IsolateOrigins,site-per-process',
]
browser = await playwright.chromium.launch(
headless=self.config.headless,
args=[
'--no-sandbox',
'--disable-blink-features=AutomationControlled',
'--disable-infobars',
'--disable-background-timer-throttling',
'--disable-popup-blocking',
'--disable-backgrounding-occluded-windows',
'--disable-renderer-backgrounding',
'--disable-window-activation',
'--disable-focus-on-load',
'--no-first-run',
'--no-default-browser-check',
'--no-startup-window',
'--window-position=0,0',
# '--window-size=1280,1000',
]
+ disable_security_args
+ self.config.extra_chromium_args,
proxy=self.config.proxy,
)
return browser
except Exception as e:
logger.error(f'Failed to initialize Playwright browser: {str(e)}')
raise

77
src/browser/custom_context.py
View File

@@ -1,10 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: context.py
import json
import logging
import os
@@ -14,7 +7,6 @@ from browser_use.browser.context import BrowserContext, BrowserContextConfig
from playwright.async_api import Browser as PlaywrightBrowser
from playwright.async_api import BrowserContext as PlaywrightBrowserContext
from .config import BrowserPersistenceConfig
logger = logging.getLogger(__name__)
@@ -25,72 +17,3 @@ class CustomBrowserContext(BrowserContext):
config: BrowserContextConfig = BrowserContextConfig()
):
super(CustomBrowserContext, self).__init__(browser=browser, config=config)
async def _create_context(self, browser: PlaywrightBrowser) -> PlaywrightBrowserContext:
"""Creates a new browser context with anti-detection measures and loads cookies if available."""
# If we have a context, return it directly
# Check if we should use existing context for persistence
if self.browser.config.chrome_instance_path and len(browser.contexts) > 0:
# Connect to existing Chrome instance instead of creating new one
context = browser.contexts[0]
else:
# Original code for creating new context
context = await browser.new_context(
viewport=self.config.browser_window_size,
no_viewport=False,
user_agent=(
"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
"(KHTML, like Gecko) Chrome/85.0.4183.102 Safari/537.36"
),
java_script_enabled=True,
bypass_csp=self.config.disable_security,
ignore_https_errors=self.config.disable_security,
record_video_dir=self.config.save_recording_path,
record_video_size=self.config.browser_window_size,
)
if self.config.trace_path:
await context.tracing.start(screenshots=True, snapshots=True, sources=True)
# Load cookies if they exist
if self.config.cookies_file and os.path.exists(self.config.cookies_file):
with open(self.config.cookies_file, "r") as f:
cookies = json.load(f)
logger.info(
f"Loaded {len(cookies)} cookies from {self.config.cookies_file}"
)
await context.add_cookies(cookies)
# Expose anti-detection scripts
await context.add_init_script(
"""
// Webdriver property
Object.defineProperty(navigator, 'webdriver', {
get: () => undefined
});
// Languages
Object.defineProperty(navigator, 'languages', {
get: () => ['en-US', 'en']
});
// Plugins
Object.defineProperty(navigator, 'plugins', {
get: () => [1, 2, 3, 4, 5]
});
// Chrome runtime
window.chrome = { runtime: {} };
// Permissions
const originalQuery = window.navigator.permissions.query;
window.navigator.permissions.query = (parameters) => (
parameters.name === 'notifications' ?
Promise.resolve({ state: Notification.permission }) :
originalQuery(parameters)
);
"""
)
return context

5
src/controller/__init__.py
View File

@@ -1,5 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: __init__.py.py

16
src/controller/custom_controller.py
View File

@@ -1,18 +1,16 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: custom_action.py
import pyperclip
from typing import Optional, Type
from pydantic import BaseModel
from browser_use.agent.views import ActionResult
from browser_use.browser.context import BrowserContext
from browser_use.controller.service import Controller
from browser_use.controller.service import Controller, DoneAction
class CustomController(Controller):
def __init__(self):
super().__init__()
def __init__(self, exclude_actions: list[str] = [],
output_model: Optional[Type[BaseModel]] = None
):
super().__init__(exclude_actions=exclude_actions, output_model=output_model)
self._register_custom_actions()
def _register_custom_actions(self):

6
src/utils/__init__.py
View File

@@ -1,6 +0,0 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: __init__.py.py

8
src/utils/default_config_settings.py
View File

@@ -11,13 +11,13 @@ def default_config():
"max_steps": 100,
"max_actions_per_step": 10,
"use_vision": True,
"tool_call_in_content": True,
"tool_calling_method": "auto",
"llm_provider": "openai",
"llm_model_name": "gpt-4o",
"llm_temperature": 1.0,
"llm_base_url": "",
"llm_api_key": "",
"use_own_browser": os.getenv("CHROME_PERSISTENT_SESSION", False),
"use_own_browser": os.getenv("CHROME_PERSISTENT_SESSION", "false").lower() == "true",
"keep_browser_open": False,
"headless": False,
"disable_security": True,
@@ -56,7 +56,7 @@ def save_current_config(*args):
"max_steps": args[1],
"max_actions_per_step": args[2],
"use_vision": args[3],
"tool_call_in_content": args[4],
"tool_calling_method": args[4],
"llm_provider": args[5],
"llm_model_name": args[6],
"llm_temperature": args[7],
@@ -86,7 +86,7 @@ def update_ui_from_config(config_file):
gr.update(value=loaded_config.get("max_steps", 100)),
gr.update(value=loaded_config.get("max_actions_per_step", 10)),
gr.update(value=loaded_config.get("use_vision", True)),
gr.update(value=loaded_config.get("tool_call_in_content", True)),
gr.update(value=loaded_config.get("tool_calling_method", True)),
gr.update(value=loaded_config.get("llm_provider", "openai")),
gr.update(value=loaded_config.get("llm_model_name", "gpt-4o")),
gr.update(value=loaded_config.get("llm_temperature", 1.0)),

35
src/utils/llm.py
View File

@@ -25,6 +25,7 @@ from langchain_core.outputs import (
LLMResult,
RunInfo,
)
from langchain_ollama import ChatOllama
from langchain_core.output_parsers.base import OutputParserLike
from langchain_core.runnables import Runnable, RunnableConfig
from langchain_core.tools import BaseTool
@@ -99,3 +100,37 @@ class DeepSeekR1ChatOpenAI(ChatOpenAI):
reasoning_content = response.choices[0].message.reasoning_content
content = response.choices[0].message.content
return AIMessage(content=content, reasoning_content=reasoning_content)
class DeepSeekR1ChatOllama(ChatOllama):
async def ainvoke(
self,
input: LanguageModelInput,
config: Optional[RunnableConfig] = None,
*,
stop: Optional[list[str]] = None,
**kwargs: Any,
) -> AIMessage:
org_ai_message = await super().ainvoke(input=input)
org_content = org_ai_message.content
reasoning_content = org_content.split("</think>")[0].replace("<think>", "")
content = org_content.split("</think>")[1]
if "**JSON Response:**" in content:
content = content.split("**JSON Response:**")[-1]
return AIMessage(content=content, reasoning_content=reasoning_content)
def invoke(
self,
input: LanguageModelInput,
config: Optional[RunnableConfig] = None,
*,
stop: Optional[list[str]] = None,
**kwargs: Any,
) -> AIMessage:
org_ai_message = super().invoke(input=input)
org_content = org_ai_message.content
reasoning_content = org_content.split("</think>")[0].replace("<think>", "")
content = org_content.split("</think>")[1]
if "**JSON Response:**" in content:
content = content.split("**JSON Response:**")[-1]
return AIMessage(content=content, reasoning_content=reasoning_content)

92
src/utils/utils.py
View File

@@ -1,9 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: utils.py
import base64
import os
import time
@@ -11,12 +5,21 @@ from pathlib import Path
from typing import Dict, Optional
from langchain_anthropic import ChatAnthropic
from langchain_mistralai import ChatMistralAI
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_ollama import ChatOllama
from langchain_openai import AzureChatOpenAI, ChatOpenAI
import gradio as gr
from .llm import DeepSeekR1ChatOpenAI
from .llm import DeepSeekR1ChatOpenAI, DeepSeekR1ChatOllama
PROVIDER_DISPLAY_NAMES = {
"openai": "OpenAI",
"azure_openai": "Azure OpenAI",
"anthropic": "Anthropic",
"deepseek": "DeepSeek",
"gemini": "Gemini"
}
def get_llm_model(provider: str, **kwargs):
"""
@@ -25,19 +28,37 @@ def get_llm_model(provider: str, **kwargs):
:param kwargs:
:return:
"""
if provider not in ["ollama"]:
env_var = "GOOGLE_API_KEY" if provider == "gemini" else f"{provider.upper()}_API_KEY"
api_key = kwargs.get("api_key", "") or os.getenv(env_var, "")
if not api_key:
handle_api_key_error(provider, env_var)
kwargs["api_key"] = api_key
if provider == "anthropic":
if not kwargs.get("base_url", ""):
base_url = "https://api.anthropic.com"
else:
base_url = kwargs.get("base_url")
return ChatAnthropic(
model_name=kwargs.get("model_name", "claude-3-5-sonnet-20240620"),
temperature=kwargs.get("temperature", 0.0),
base_url=base_url,
api_key=api_key,
)
elif provider == 'mistral':
if not kwargs.get("base_url", ""):
base_url = os.getenv("MISTRAL_ENDPOINT", "https://api.mistral.ai/v1")
else:
base_url = kwargs.get("base_url")
if not kwargs.get("api_key", ""):
api_key = os.getenv("ANTHROPIC_API_KEY", "")
api_key = os.getenv("MISTRAL_API_KEY", "")
else:
api_key = kwargs.get("api_key")
return ChatAnthropic(
model_name=kwargs.get("model_name", "claude-3-5-sonnet-20240620"),
return ChatMistralAI(
model=kwargs.get("model_name", "mistral-large-latest"),
temperature=kwargs.get("temperature", 0.0),
base_url=base_url,
api_key=api_key,
@@ -48,11 +69,6 @@ def get_llm_model(provider: str, **kwargs):
else:
base_url = kwargs.get("base_url")
if not kwargs.get("api_key", ""):
api_key = os.getenv("OPENAI_API_KEY", "")
else:
api_key = kwargs.get("api_key")
return ChatOpenAI(
model=kwargs.get("model_name", "gpt-4o"),
temperature=kwargs.get("temperature", 0.0),
@@ -65,11 +81,6 @@ def get_llm_model(provider: str, **kwargs):
else:
base_url = kwargs.get("base_url")
if not kwargs.get("api_key", ""):
api_key = os.getenv("DEEPSEEK_API_KEY", "")
else:
api_key = kwargs.get("api_key")
if kwargs.get("model_name", "deepseek-chat") == "deepseek-reasoner":
return DeepSeekR1ChatOpenAI(
model=kwargs.get("model_name", "deepseek-reasoner"),
@@ -85,31 +96,37 @@ def get_llm_model(provider: str, **kwargs):
api_key=api_key,
)
elif provider == "gemini":
if not kwargs.get("api_key", ""):
api_key = os.getenv("GOOGLE_API_KEY", "")
else:
api_key = kwargs.get("api_key")
return ChatGoogleGenerativeAI(
model=kwargs.get("model_name", "gemini-2.0-flash-exp"),
temperature=kwargs.get("temperature", 0.0),
google_api_key=api_key,
)
elif provider == "ollama":
if not kwargs.get("base_url", ""):
base_url = os.getenv("OLLAMA_ENDPOINT", "http://localhost:11434")
else:
base_url = kwargs.get("base_url")
if "deepseek-r1" in kwargs.get("model_name", "qwen2.5:7b"):
return DeepSeekR1ChatOllama(
model=kwargs.get("model_name", "deepseek-r1:14b"),
temperature=kwargs.get("temperature", 0.0),
num_ctx=kwargs.get("num_ctx", 32000),
base_url=base_url,
)
else:
return ChatOllama(
model=kwargs.get("model_name", "qwen2.5:7b"),
temperature=kwargs.get("temperature", 0.0),
num_ctx=kwargs.get("num_ctx", 32000),
base_url=kwargs.get("base_url", "http://localhost:11434"),
num_predict=kwargs.get("num_predict", 1024),
base_url=base_url,
)
elif provider == "azure_openai":
if not kwargs.get("base_url", ""):
base_url = os.getenv("AZURE_OPENAI_ENDPOINT", "")
else:
base_url = kwargs.get("base_url")
if not kwargs.get("api_key", ""):
api_key = os.getenv("AZURE_OPENAI_API_KEY", "")
else:
api_key = kwargs.get("api_key")
return AzureChatOpenAI(
model=kwargs.get("model_name", "gpt-4o"),
temperature=kwargs.get("temperature", 0.0),
@@ -123,11 +140,12 @@ def get_llm_model(provider: str, **kwargs):
# Predefined model names for common providers
model_names = {
"anthropic": ["claude-3-5-sonnet-20240620", "claude-3-opus-20240229"],
"openai": ["gpt-4o", "gpt-4", "gpt-3.5-turbo"],
"openai": ["gpt-4o", "gpt-4", "gpt-3.5-turbo", "o3-mini"],
"deepseek": ["deepseek-chat", "deepseek-reasoner"],
"gemini": ["gemini-2.0-flash-exp", "gemini-2.0-flash-thinking-exp", "gemini-1.5-flash-latest", "gemini-1.5-flash-8b-latest", "gemini-2.0-flash-thinking-exp-1219" ],
"ollama": ["qwen2.5:7b", "llama2:7b"],
"azure_openai": ["gpt-4o", "gpt-4", "gpt-3.5-turbo"]
"ollama": ["qwen2.5:7b", "llama2:7b", "deepseek-r1:14b", "deepseek-r1:32b"],
"azure_openai": ["gpt-4o", "gpt-4", "gpt-3.5-turbo"],
"mistral": ["pixtral-large-latest", "mistral-large-latest", "mistral-small-latest", "ministral-8b-latest"]
}
# Callback to update the model name dropdown based on the selected provider
@@ -147,6 +165,16 @@ def update_model_dropdown(llm_provider, api_key=None, base_url=None):
else:
return gr.Dropdown(choices=[], value="", interactive=True, allow_custom_value=True)
def handle_api_key_error(provider: str, env_var: str):
"""
Handles the missing API key error by raising a gr.Error with a clear message.
"""
provider_display = PROVIDER_DISPLAY_NAMES.get(provider, provider.upper())
raise gr.Error(
f"💥 {provider_display} API key not found! 🔑 Please set the "
f"`{env_var}` environment variable or provide it in the UI."
)
def encode_image(img_path):
if not img_path:
return None

205
tests/test_browser_use.py
View File

@@ -1,8 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @ProjectName: browser-use-webui
# @FileName: test_browser_use.py
import pdb
from dotenv import load_dotenv
@@ -37,15 +32,27 @@ async def test_browser_use_org():
# api_key=os.getenv("AZURE_OPENAI_API_KEY", ""),
# )
# llm = utils.get_llm_model(
# provider="deepseek",
# model_name="deepseek-chat",
# temperature=0.8
# )
llm = utils.get_llm_model(
provider="deepseek",
model_name="deepseek-chat",
temperature=0.8
provider="ollama", model_name="deepseek-r1:14b", temperature=0.5
)
window_w, window_h = 1920, 1080
use_vision = False
use_own_browser = False
if use_own_browser:
chrome_path = os.getenv("CHROME_PATH", None)
if chrome_path == "":
chrome_path = None
else:
chrome_path = None
tool_calling_method = "json_schema" # setting to json_schema when using ollma
browser = Browser(
config=BrowserConfig(
@@ -69,7 +76,8 @@ async def test_browser_use_org():
task="go to google.com and type 'OpenAI' click search and give me the first url",
llm=llm,
browser_context=browser_context,
use_vision=use_vision
use_vision=use_vision,
tool_calling_method=tool_calling_method
)
history: AgentHistoryList = await agent.run(max_steps=10)
@@ -95,7 +103,7 @@ async def test_browser_use_custom():
from playwright.async_api import async_playwright
from src.agent.custom_agent import CustomAgent
from src.agent.custom_prompts import CustomSystemPrompt
from src.agent.custom_prompts import CustomSystemPrompt, CustomAgentMessagePrompt
from src.browser.custom_browser import CustomBrowser
from src.browser.custom_context import BrowserContextConfig
from src.controller.custom_controller import CustomController
@@ -103,143 +111,21 @@ async def test_browser_use_custom():
window_w, window_h = 1920, 1080
# llm = utils.get_llm_model(
# provider="azure_openai",
# provider="openai",
# model_name="gpt-4o",
# temperature=0.8,
# base_url=os.getenv("AZURE_OPENAI_ENDPOINT", ""),
# api_key=os.getenv("AZURE_OPENAI_API_KEY", ""),
# base_url=os.getenv("OPENAI_ENDPOINT", ""),
# api_key=os.getenv("OPENAI_API_KEY", ""),
# )
llm = utils.get_llm_model(
provider="gemini",
model_name="gemini-2.0-flash-exp",
temperature=1.0,
api_key=os.getenv("GOOGLE_API_KEY", "")
provider="azure_openai",
model_name="gpt-4o",
temperature=0.8,
base_url=os.getenv("AZURE_OPENAI_ENDPOINT", ""),
api_key=os.getenv("AZURE_OPENAI_API_KEY", ""),
)
# llm = utils.get_llm_model(
# provider="deepseek",
# model_name="deepseek-chat",
# temperature=0.8
# )
# llm = utils.get_llm_model(
# provider="ollama", model_name="qwen2.5:7b", temperature=0.8
# )
controller = CustomController()
use_own_browser = False
disable_security = True
use_vision = True # Set to False when using DeepSeek
tool_call_in_content = True # Set to True when using Ollama
max_actions_per_step = 1
playwright = None
browser_context_ = None
try:
if use_own_browser:
playwright = await async_playwright().start()
chrome_exe = os.getenv("CHROME_PATH", "")
chrome_use_data = os.getenv("CHROME_USER_DATA", "")
browser_context_ = await playwright.chromium.launch_persistent_context(
user_data_dir=chrome_use_data,
executable_path=chrome_exe,
no_viewport=False,
headless=False, # 保持浏览器窗口可见
user_agent=(
"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
"(KHTML, like Gecko) Chrome/85.0.4183.102 Safari/537.36"
),
java_script_enabled=True,
bypass_csp=disable_security,
ignore_https_errors=disable_security,
record_video_dir="./tmp/record_videos",
record_video_size={"width": window_w, "height": window_h},
)
else:
browser_context_ = None
browser = CustomBrowser(
config=BrowserConfig(
headless=False,
disable_security=True,
extra_chromium_args=[f"--window-size={window_w},{window_h}"],
)
)
async with await browser.new_context(
config=BrowserContextConfig(
trace_path="./tmp/result_processing",
save_recording_path="./tmp/record_videos",
no_viewport=False,
browser_window_size=BrowserContextWindowSize(
width=window_w, height=window_h
),
),
context=browser_context_,
) as browser_context:
agent = CustomAgent(
task="go to google.com and type 'OpenAI' click search and give me the first url",
add_infos="", # some hints for llm to complete the task
llm=llm,
browser_context=browser_context,
controller=controller,
system_prompt_class=CustomSystemPrompt,
use_vision=use_vision,
tool_call_in_content=tool_call_in_content,
max_actions_per_step=max_actions_per_step
)
history: AgentHistoryList = await agent.run(max_steps=10)
print("Final Result:")
pprint(history.final_result(), indent=4)
print("\nErrors:")
pprint(history.errors(), indent=4)
# e.g. xPaths the model clicked on
print("\nModel Outputs:")
pprint(history.model_actions(), indent=4)
print("\nThoughts:")
pprint(history.model_thoughts(), indent=4)
# close browser
except Exception:
import traceback
traceback.print_exc()
finally:
# 显式关闭持久化上下文
if browser_context_:
await browser_context_.close()
# 关闭 Playwright 对象
if playwright:
await playwright.stop()
await browser.close()
async def test_browser_use_custom_v2():
from browser_use.browser.context import BrowserContextWindowSize
from browser_use.browser.browser import BrowserConfig
from playwright.async_api import async_playwright
from src.agent.custom_agent import CustomAgent
from src.agent.custom_prompts import CustomSystemPrompt
from src.browser.custom_browser import CustomBrowser
from src.browser.custom_context import BrowserContextConfig
from src.controller.custom_controller import CustomController
window_w, window_h = 1920, 1080
# llm = utils.get_llm_model(
# provider="azure_openai",
# model_name="gpt-4o",
# temperature=0.8,
# base_url=os.getenv("AZURE_OPENAI_ENDPOINT", ""),
# api_key=os.getenv("AZURE_OPENAI_API_KEY", ""),
# )
# llm = utils.get_llm_model(
# provider="gemini",
# model_name="gemini-2.0-flash-exp",
@@ -247,31 +133,45 @@ async def test_browser_use_custom_v2():
# api_key=os.getenv("GOOGLE_API_KEY", "")
# )
llm = utils.get_llm_model(
provider="deepseek",
model_name="deepseek-reasoner",
temperature=0.8
)
# llm = utils.get_llm_model(
# provider="deepseek",
# model_name="deepseek-reasoner",
# temperature=0.8
# )
# llm = utils.get_llm_model(
# provider="deepseek",
# model_name="deepseek-chat",
# temperature=0.8
# )
# llm = utils.get_llm_model(
# provider="ollama", model_name="qwen2.5:7b", temperature=0.5
# )
# llm = utils.get_llm_model(
# provider="ollama", model_name="deepseek-r1:14b", temperature=0.5
# )
controller = CustomController()
use_own_browser = False
use_own_browser = True
disable_security = True
use_vision = False # Set to False when using DeepSeek
tool_call_in_content = True # Set to True when using Ollama
max_actions_per_step = 1
playwright = None
browser = None
browser_context = None
try:
extra_chromium_args = [f"--window-size={window_w},{window_h}"]
if use_own_browser:
chrome_path = os.getenv("CHROME_PATH", None)
if chrome_path == "":
chrome_path = None
chrome_user_data = os.getenv("CHROME_USER_DATA", None)
if chrome_user_data:
extra_chromium_args += [f"--user-data-dir={chrome_user_data}"]
else:
chrome_path = None
browser = CustomBrowser(
@@ -279,7 +179,7 @@ async def test_browser_use_custom_v2():
headless=False,
disable_security=disable_security,
chrome_instance_path=chrome_path,
extra_chromium_args=[f"--window-size={window_w},{window_h}"],
extra_chromium_args=extra_chromium_args,
)
)
browser_context = await browser.new_context(
@@ -293,18 +193,18 @@ async def test_browser_use_custom_v2():
)
)
agent = CustomAgent(
task="go to google.com and type 'OpenAI' click search and give me the first url",
task="Search 'Nvidia' and give me the first url",
add_infos="", # some hints for llm to complete the task
llm=llm,
browser=browser,
browser_context=browser_context,
controller=controller,
system_prompt_class=CustomSystemPrompt,
agent_prompt_class=CustomAgentMessagePrompt,
use_vision=use_vision,
tool_call_in_content=tool_call_in_content,
max_actions_per_step=max_actions_per_step
)
history: AgentHistoryList = await agent.run(max_steps=10)
history: AgentHistoryList = await agent.run(max_steps=100)
print("Final Result:")
pprint(history.final_result(), indent=4)
@@ -336,5 +236,4 @@ async def test_browser_use_custom_v2():
if __name__ == "__main__":
# asyncio.run(test_browser_use_org())
# asyncio.run(test_browser_use_custom())
asyncio.run(test_browser_use_custom_v2())
asyncio.run(test_browser_use_custom())

209
tests/test_llm_api.py
View File

@@ -1,13 +1,10 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: test_llm_api.py
import os
import pdb
from dataclasses import dataclass
from dotenv import load_dotenv
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_ollama import ChatOllama
load_dotenv()
@@ -15,145 +12,121 @@ import sys
sys.path.append(".")
@dataclass
class LLMConfig:
provider: str
model_name: str
temperature: float = 0.8
base_url: str = None
api_key: str = None
def test_openai_model():
from langchain_core.messages import HumanMessage
def create_message_content(text, image_path=None):
content = [{"type": "text", "text": text}]
if image_path:
from src.utils import utils
llm = utils.get_llm_model(
provider="openai",
model_name="gpt-4o",
temperature=0.8,
base_url=os.getenv("OPENAI_ENDPOINT", ""),
api_key=os.getenv("OPENAI_API_KEY", "")
)
image_path = "assets/examples/test.png"
image_data = utils.encode_image(image_path)
message = HumanMessage(
content=[
{"type": "text", "text": "describe this image"},
{
content.append({
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{image_data}"},
},
]
)
ai_msg = llm.invoke([message])
print(ai_msg.content)
"image_url": {"url": f"data:image/jpeg;base64,{image_data}"}
})
return content
def test_gemini_model():
# you need to enable your api key first: https://ai.google.dev/palm_docs/oauth_quickstart
from langchain_core.messages import HumanMessage
def get_env_value(key, provider):
env_mappings = {
"openai": {"api_key": "OPENAI_API_KEY", "base_url": "OPENAI_ENDPOINT"},
"azure_openai": {"api_key": "AZURE_OPENAI_API_KEY", "base_url": "AZURE_OPENAI_ENDPOINT"},
"gemini": {"api_key": "GOOGLE_API_KEY"},
"deepseek": {"api_key": "DEEPSEEK_API_KEY", "base_url": "DEEPSEEK_ENDPOINT"},
"mistral": {"api_key": "MISTRAL_API_KEY", "base_url": "MISTRAL_ENDPOINT"},
}
if provider in env_mappings and key in env_mappings[provider]:
return os.getenv(env_mappings[provider][key], "")
return ""
def test_llm(config, query, image_path=None, system_message=None):
from src.utils import utils
llm = utils.get_llm_model(
provider="gemini",
model_name="gemini-2.0-flash-exp",
temperature=0.8,
api_key=os.getenv("GOOGLE_API_KEY", "")
)
# Special handling for Ollama-based models
if config.provider == "ollama":
if "deepseek-r1" in config.model_name:
from src.utils.llm import DeepSeekR1ChatOllama
llm = DeepSeekR1ChatOllama(model=config.model_name)
else:
llm = ChatOllama(model=config.model_name)
image_path = "assets/examples/test.png"
image_data = utils.encode_image(image_path)
message = HumanMessage(
content=[
{"type": "text", "text": "describe this image"},
{
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{image_data}"},
},
]
)
ai_msg = llm.invoke([message])
ai_msg = llm.invoke(query)
print(ai_msg.content)
if "deepseek-r1" in config.model_name:
pdb.set_trace()
return
def test_azure_openai_model():
from langchain_core.messages import HumanMessage
from src.utils import utils
# For other providers, use the standard configuration
llm = utils.get_llm_model(
provider="azure_openai",
model_name="gpt-4o",
temperature=0.8,
base_url=os.getenv("AZURE_OPENAI_ENDPOINT", ""),
api_key=os.getenv("AZURE_OPENAI_API_KEY", "")
provider=config.provider,
model_name=config.model_name,
temperature=config.temperature,
base_url=config.base_url or get_env_value("base_url", config.provider),
api_key=config.api_key or get_env_value("api_key", config.provider)
)
image_path = "assets/examples/test.png"
image_data = utils.encode_image(image_path)
message = HumanMessage(
content=[
{"type": "text", "text": "describe this image"},
{
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{image_data}"},
},
]
)
ai_msg = llm.invoke([message])
print(ai_msg.content)
def test_deepseek_model():
from langchain_core.messages import HumanMessage
from src.utils import utils
llm = utils.get_llm_model(
provider="deepseek",
model_name="deepseek-chat",
temperature=0.8,
base_url=os.getenv("DEEPSEEK_ENDPOINT", ""),
api_key=os.getenv("DEEPSEEK_API_KEY", "")
)
message = HumanMessage(
content=[
{"type": "text", "text": "who are you?"}
]
)
ai_msg = llm.invoke([message])
print(ai_msg.content)
def test_deepseek_r1_model():
from langchain_core.messages import HumanMessage, SystemMessage, AIMessage
from src.utils import utils
llm = utils.get_llm_model(
provider="deepseek",
model_name="deepseek-reasoner",
temperature=0.8,
base_url=os.getenv("DEEPSEEK_ENDPOINT", ""),
api_key=os.getenv("DEEPSEEK_API_KEY", "")
)
# Prepare messages for non-Ollama models
messages = []
sys_message = SystemMessage(
content=[{"type": "text", "text": "you are a helpful AI assistant"}]
)
messages.append(sys_message)
user_message = HumanMessage(
content=[
{"type": "text", "text": "9.11 and 9.8, which is greater?"}
]
)
messages.append(user_message)
if system_message:
messages.append(SystemMessage(content=create_message_content(system_message)))
messages.append(HumanMessage(content=create_message_content(query, image_path)))
ai_msg = llm.invoke(messages)
# Handle different response types
if hasattr(ai_msg, "reasoning_content"):
print(ai_msg.reasoning_content)
print(ai_msg.content)
if config.provider == "deepseek" and "deepseek-reasoner" in config.model_name:
print(llm.model_name)
pdb.set_trace()
def test_openai_model():
config = LLMConfig(provider="openai", model_name="gpt-4o")
test_llm(config, "Describe this image", "assets/examples/test.png")
def test_gemini_model():
# Enable your API key first if you haven't: https://ai.google.dev/palm_docs/oauth_quickstart
config = LLMConfig(provider="gemini", model_name="gemini-2.0-flash-exp")
test_llm(config, "Describe this image", "assets/examples/test.png")
def test_azure_openai_model():
config = LLMConfig(provider="azure_openai", model_name="gpt-4o")
test_llm(config, "Describe this image", "assets/examples/test.png")
def test_deepseek_model():
config = LLMConfig(provider="deepseek", model_name="deepseek-chat")
test_llm(config, "Who are you?")
def test_deepseek_r1_model():
config = LLMConfig(provider="deepseek", model_name="deepseek-reasoner")
test_llm(config, "Which is greater, 9.11 or 9.8?", system_message="You are a helpful AI assistant.")
def test_ollama_model():
from langchain_ollama import ChatOllama
config = LLMConfig(provider="ollama", model_name="qwen2.5:7b")
test_llm(config, "Sing a ballad of LangChain.")
llm = ChatOllama(model="qwen2.5:7b")
ai_msg = llm.invoke("Sing a ballad of LangChain.")
print(ai_msg.content)
def test_deepseek_r1_ollama_model():
config = LLMConfig(provider="ollama", model_name="deepseek-r1:14b")
test_llm(config, "How many 'r's are in the word 'strawberry'?")
def test_mistral_model():
config = LLMConfig(provider="mistral", model_name="pixtral-large-latest")
test_llm(config, "Describe this image", "assets/examples/test.png")
if __name__ == '__main__':
if __name__ == "__main__":
# test_openai_model()
# test_gemini_model()
# test_azure_openai_model()
#test_deepseek_model()
# test_ollama_model()
test_deepseek_r1_model()
# test_deepseek_r1_model()
# test_deepseek_r1_ollama_model()
test_mistral_model()

6
tests/test_playwright.py
View File

@@ -1,9 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/2
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: test_playwright.py
import pdb
from dotenv import load_dotenv

89
webui.py
View File

@@ -1,10 +1,3 @@
# -*- coding: utf-8 -*-
# @Time : 2025/1/1
# @Author : wenshao
# @Email : wenshaoguo1026@gmail.com
# @Project : browser-use-webui
# @FileName: webui.py
import pdb
import logging
@@ -28,22 +21,20 @@ from browser_use.browser.context import (
BrowserContextConfig,
BrowserContextWindowSize,
)
from langchain_ollama import ChatOllama
from playwright.async_api import async_playwright
from src.utils.agent_state import AgentState
from src.utils import utils
from src.agent.custom_agent import CustomAgent
from src.browser.custom_browser import CustomBrowser
from src.agent.custom_prompts import CustomSystemPrompt
from src.browser.config import BrowserPersistenceConfig
from src.agent.custom_prompts import CustomSystemPrompt, CustomAgentMessagePrompt
from src.browser.custom_context import BrowserContextConfig, CustomBrowserContext
from src.controller.custom_controller import CustomController
from gradio.themes import Citrus, Default, Glass, Monochrome, Ocean, Origin, Soft, Base
from src.utils.default_config_settings import default_config, load_config_from_file, save_config_to_file, save_current_config, update_ui_from_config
from src.utils.utils import update_model_dropdown, get_latest_files, capture_screenshot
from dotenv import load_dotenv
load_dotenv()
# Global variables for persistence
_global_browser = None
@@ -101,7 +92,7 @@ async def run_browser_agent(
max_steps,
use_vision,
max_actions_per_step,
tool_call_in_content
tool_calling_method
):
global _global_agent_state
_global_agent_state.clear_stop() # Clear any previous stop requests
@@ -147,7 +138,7 @@ async def run_browser_agent(
max_steps=max_steps,
use_vision=use_vision,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content
tool_calling_method=tool_calling_method
)
elif agent_type == "custom":
final_result, errors, model_actions, model_thoughts, trace_file, history_file = await run_custom_agent(
@@ -166,7 +157,7 @@ async def run_browser_agent(
max_steps=max_steps,
use_vision=use_vision,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content
tool_calling_method=tool_calling_method
)
else:
raise ValueError(f"Invalid agent type: {agent_type}")
@@ -193,6 +184,9 @@ async def run_browser_agent(
gr.update(interactive=True) # Re-enable run button
)
except gr.Error:
raise
except Exception as e:
import traceback
traceback.print_exc()
@@ -225,7 +219,7 @@ async def run_org_agent(
max_steps,
use_vision,
max_actions_per_step,
tool_call_in_content
tool_calling_method
):
try:
global _global_browser, _global_browser_context, _global_agent_state
@@ -233,10 +227,14 @@ async def run_org_agent(
# Clear any previous stop request
_global_agent_state.clear_stop()
extra_chromium_args = [f"--window-size={window_w},{window_h}"]
if use_own_browser:
chrome_path = os.getenv("CHROME_PATH", None)
if chrome_path == "":
chrome_path = None
chrome_user_data = os.getenv("CHROME_USER_DATA", None)
if chrome_user_data:
extra_chromium_args += [f"--user-data-dir={chrome_user_data}"]
else:
chrome_path = None
@@ -246,7 +244,7 @@ async def run_org_agent(
headless=headless,
disable_security=disable_security,
chrome_instance_path=chrome_path,
extra_chromium_args=[f"--window-size={window_w},{window_h}"],
extra_chromium_args=extra_chromium_args,
)
)
@@ -269,7 +267,7 @@ async def run_org_agent(
browser=_global_browser,
browser_context=_global_browser_context,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content
tool_calling_method=tool_calling_method
)
history = await agent.run(max_steps=max_steps)
@@ -316,7 +314,7 @@ async def run_custom_agent(
max_steps,
use_vision,
max_actions_per_step,
tool_call_in_content
tool_calling_method
):
try:
global _global_browser, _global_browser_context, _global_agent_state
@@ -324,10 +322,14 @@ async def run_custom_agent(
# Clear any previous stop request
_global_agent_state.clear_stop()
extra_chromium_args = [f"--window-size={window_w},{window_h}"]
if use_own_browser:
chrome_path = os.getenv("CHROME_PATH", None)
if chrome_path == "":
chrome_path = None
chrome_user_data = os.getenv("CHROME_USER_DATA", None)
if chrome_user_data:
extra_chromium_args += [f"--user-data-dir={chrome_user_data}"]
else:
chrome_path = None
@@ -340,7 +342,7 @@ async def run_custom_agent(
headless=headless,
disable_security=disable_security,
chrome_instance_path=chrome_path,
extra_chromium_args=[f"--window-size={window_w},{window_h}"],
extra_chromium_args=extra_chromium_args,
)
)
@@ -366,9 +368,10 @@ async def run_custom_agent(
browser_context=_global_browser_context,
controller=controller,
system_prompt_class=CustomSystemPrompt,
agent_prompt_class=CustomAgentMessagePrompt,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content,
agent_state=_global_agent_state
agent_state=_global_agent_state,
tool_calling_method=tool_calling_method
)
history = await agent.run(max_steps=max_steps)
@@ -421,7 +424,7 @@ async def run_with_stream(
max_steps,
use_vision,
max_actions_per_step,
tool_call_in_content
tool_calling_method
):
global _global_agent_state
stream_vw = 80
@@ -449,7 +452,7 @@ async def run_with_stream(
max_steps=max_steps,
use_vision=use_vision,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content
tool_calling_method=tool_calling_method
)
# Add HTML content at the start of the result array
html_content = f"<h1 style='width:{stream_vw}vw; height:{stream_vh}vh'>Using browser...</h1>"
@@ -481,7 +484,7 @@ async def run_with_stream(
max_steps=max_steps,
use_vision=use_vision,
max_actions_per_step=max_actions_per_step,
tool_call_in_content=tool_call_in_content
tool_calling_method=tool_calling_method
)
)
@@ -535,6 +538,12 @@ async def run_with_stream(
try:
result = await agent_task
final_result, errors, model_actions, model_thoughts, latest_videos, trace, history_file, stop_button, run_button = result
except gr.Error:
final_result = ""
model_actions = ""
model_thoughts = ""
latest_videos = trace = history_file = None
except Exception as e:
errors = f"Agent error: {str(e)}"
@@ -607,18 +616,8 @@ def create_ui(config, theme_name="Ocean"):
}
"""
js = """
function refresh() {
const url = new URL(window.location);
if (url.searchParams.get('__theme') !== 'dark') {
url.searchParams.set('__theme', 'dark');
window.location.href = url.href;
}
}
"""
with gr.Blocks(
title="Browser Use WebUI", theme=theme_map[theme_name], css=css, js=js
title="Browser Use WebUI", theme=theme_map[theme_name], css=css
) as demo:
with gr.Row():
gr.Markdown(
@@ -638,6 +637,7 @@ def create_ui(config, theme_name="Ocean"):
value=config['agent_type'],
info="Select the type of agent to use",
)
with gr.Column():
max_steps = gr.Slider(
minimum=1,
maximum=200,
@@ -654,15 +654,20 @@ def create_ui(config, theme_name="Ocean"):
label="Max Actions per Step",
info="Maximum number of actions the agent will take per step",
)
with gr.Column():
use_vision = gr.Checkbox(
label="Use Vision",
value=config['use_vision'],
info="Enable visual processing capabilities",
)
tool_call_in_content = gr.Checkbox(
label="Use Tool Calls in Content",
value=config['tool_call_in_content'],
info="Enable Tool Calls in content",
tool_calling_method = gr.Dropdown(
label="Tool Calling Method",
value=config['tool_calling_method'],
interactive=True,
allow_custom_value=True, # Allow users to input custom model names
choices=["auto", "json_schema", "function_calling"],
info="Tool Calls Funtion Name",
visible=False
)
with gr.TabItem("🔧 LLM Configuration", id=2):
@@ -813,7 +818,7 @@ def create_ui(config, theme_name="Ocean"):
fn=update_ui_from_config,
inputs=[config_file_input],
outputs=[
agent_type, max_steps, max_actions_per_step, use_vision, tool_call_in_content,
agent_type, max_steps, max_actions_per_step, use_vision, tool_calling_method,
llm_provider, llm_model_name, llm_temperature, llm_base_url, llm_api_key,
use_own_browser, keep_browser_open, headless, disable_security, enable_recording,
window_w, window_h, save_recording_path, save_trace_path, save_agent_history_path,
@@ -824,7 +829,7 @@ def create_ui(config, theme_name="Ocean"):
save_config_button.click(
fn=save_current_config,
inputs=[
agent_type, max_steps, max_actions_per_step, use_vision, tool_call_in_content,
agent_type, max_steps, max_actions_per_step, use_vision, tool_calling_method,
llm_provider, llm_model_name, llm_temperature, llm_base_url, llm_api_key,
use_own_browser, keep_browser_open, headless, disable_security,
enable_recording, window_w, window_h, save_recording_path, save_trace_path,
@@ -876,7 +881,7 @@ def create_ui(config, theme_name="Ocean"):
agent_type, llm_provider, llm_model_name, llm_temperature, llm_base_url, llm_api_key,
use_own_browser, keep_browser_open, headless, disable_security, window_w, window_h,
save_recording_path, save_agent_history_path, save_trace_path, # Include the new path
enable_recording, task, add_infos, max_steps, use_vision, max_actions_per_step, tool_call_in_content
enable_recording, task, add_infos, max_steps, use_vision, max_actions_per_step, tool_calling_method
],
outputs=[
browser_view, # Browser view