阿咕噜 AI 实战手册

Appearance

卷不动了，真的卷不动了，OWL也来砸Manus的场子了 ​

昨天刚介绍完艾莉同学，今天就又来个OWL，我出教程都有点扛不住了，每天写稿都要写到凌晨以后，卷不动，实在是卷不动了。好想找个AI帮我写，各位看官有推荐的吗，介绍给我试试！！！

一、OWL是啥？ ​

OWL跟Manus一样，都是一个的多智能体协作框架，通过利用动态智能体交互，实现跨领域，自然、高效且稳健的自动执行任务。

根据官网描述：OWL 在 GAIA 基准测试中取得 58.18 平均分，在开源框架中排名 🏅️ #1！

这么来说还是蛮牛逼的，具体架构如下：

二、入门指南 ​

我是基于Windows11来操作的，如果想少走弯路，一把在自己电脑上跑起来，请务必跟随我的步骤来！

1. 环境准备 ​

官网推荐用UV，之前有帖子讲过怎么安装，不懂的可以上官网：https://docs.astral.sh/uv ，由于安装都需要有外网，所以建议最好做全局代理，然后终端也加个代理，方便后面安装python，然后在你的终端执行如下命令：curl ipinfo.io ，出现外网地址则表示成功。

bash
# 克隆 GitHub 仓库
git clone https://github.com/camel-ai/owl.git

# 进入项目目录
cd owl

# 创建虚拟环境并安装依赖
# 我们支持使用 Python 3.10、3.11、3.12
# 注意，这里最好要用3.10的版本，我之前用的3.12结果出现了些奇怪的问题
uv venv .venv --python=3.10

# 激活虚拟环境
.venv\Scripts\activate

# 安装 CAMEL 及其所有依赖（python3.12的版本在这里安装依赖可能会出错）
uv pip install -e .

2. 修改配置 ​

在owl目录下复制.env.example，并重命名为.env。

我这边演示用是硅基流动的API，注册地址：https://cloud.siliconflow.com/i/0hq76Lbt ，只需要如下配置即可。

# SILICONFLOW API
SILICONFLOW_API_KEY='sk-********z'

# Google Search API (https://coda.io/@jon-dallas/google-image-search-pack-example/search-engine-id-and-google-api-key-3)
GOOGLE_API_KEY='AI********A'
SEARCH_ENGINE_ID='a********0'

# Chunkr API (https://chunkr.ai/)
CHUNKR_API_KEY='ch_********'

# Firecrawl API (https://www.firecrawl.dev/)
FIRECRAWL_API_KEY='fc-********'
#FIRECRAWL_API_URL="https://api.firecrawl.dev"

接下来我们需要修改代码，官方其实是支持供硅基流动的API的，可以看到官方源码里面是包含了硅基流动的API地址的，所以我们的配置文件里面不用填写URL，只需要配置API_KEY就行。

官方的例子：https://github.com/camel-ai/camel/blob/master/examples/models/siliconflow_model_example.py

由于官方的示例里面没有硅基流动，所以需要们自己加一个。在examples里复制run_mini.py并重命名为run_silicon_flow.py

run_silicon_flow.py

python
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
import sys
from dotenv import load_dotenv

from camel.models import ModelFactory
from camel.toolkits import (
    SearchToolkit,
    BrowserToolkit,
    FileWriteToolkit,
)
from camel.types import ModelPlatformType, ModelType
from camel.logger import set_log_level

from owl.utils import run_society

from camel.societies import RolePlaying

import pathlib

base_dir = pathlib.Path(__file__).parent.parent
env_path = base_dir / "owl" / ".env"
load_dotenv(dotenv_path=str(env_path))

set_log_level(level="DEBUG")


def construct_society(question: str) -> RolePlaying:
    r"""Construct a society of agents based on the given question.

    Args:
        question (str): The task or question to be addressed by the society.

    Returns:
        RolePlaying: A configured society of agents ready to address the
            question.
    """

    # Create models for different components
    models = {
        "user": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
        "assistant": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
        "browsing": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/Qwen2.5-VL-72B-Instruct",
            model_config_dict={"temperature": 0},
        ),
        "planning": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
    }


    # Configure toolkits
    tools = [
        *BrowserToolkit(
            headless=False,  # Set to True for headless mode (e.g., on remote servers)
            web_agent_model=models["browsing"],
            planning_agent_model=models["planning"],
        ).get_tools(),
        SearchToolkit().search_duckduckgo,
        SearchToolkit().search_wiki,
        *FileWriteToolkit(output_dir="./").get_tools(),
    ]

    # Configure agent roles and parameters
    user_agent_kwargs = {"model": models["user"]}
    assistant_agent_kwargs = {"model": models["assistant"], "tools": tools}

    # Configure task parameters
    task_kwargs = {
        "task_prompt": question,
        "with_task_specify": False,
    }

    # Create and return the society
    society = RolePlaying(
        **task_kwargs,
        user_role_name="user",
        user_agent_kwargs=user_agent_kwargs,
        assistant_role_name="assistant",
        assistant_agent_kwargs=assistant_agent_kwargs,
    )

    return society


def main():
    r"""Main function to run the OWL system with an example question."""
    # Default research question
    default_task = "统计豆瓣电影top10的电影名称."

    # Override default task if command line argument is provided
    task = sys.argv[1] if len(sys.argv) > 1 else default_task

    # Construct and run the society
    society = construct_society(task)
    answer, chat_history, token_count = run_society(society)

    # Output the result
    print(f"\033[94mAnswer: {answer}\033[0m")


if __name__ == "__main__":
    main()

这些配置完成后就可以在终端执行该命令：python examples\run_silicon_flow.py ，可以看到以下结果，只是不怎么准确，因为有些工具还没配置。

image-20250325000947377

3. 添加其他增强工具 ​

细心的朋友可能会发现run.py和run_mini.py的tools是不一样的，这也就是为什么mini生成的结果总是不一样，因为他少了很多东西。

现在我们来调整一下，将之前的代码改成如下代码：

run_silicon_flow.py

python
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
import sys
import pathlib
from dotenv import load_dotenv
from camel.models import ModelFactory
from camel.toolkits import (
    AudioAnalysisToolkit,
    CodeExecutionToolkit,
    ExcelToolkit,
    ImageAnalysisToolkit,
    SearchToolkit,
    VideoAnalysisToolkit,
    BrowserToolkit,
    FileWriteToolkit,
)
from camel.types import ModelPlatformType, ModelType
from camel.logger import set_log_level
from camel.societies import RolePlaying

from owl.utils import run_society, DocumentProcessingToolkit

base_dir = pathlib.Path(__file__).parent.parent
env_path = base_dir / "owl" / ".env"
load_dotenv(dotenv_path=str(env_path))

set_log_level(level="DEBUG")


def construct_society(question: str) -> RolePlaying:
    r"""Construct a society of agents based on the given question.

    Args:
        question (str): The task or question to be addressed by the society.

    Returns:
        RolePlaying: A configured society of agents ready to address the question.
    """

    # Create models for different components
    models = {
        "user": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
        "assistant": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
        "browsing": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/Qwen2.5-VL-72B-Instruct",
            model_config_dict={"temperature": 0},
        ),
        "planning": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/QwQ-32B",
            model_config_dict={"temperature": 0},
        ),
        "video": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/Qwen2.5-VL-72B-Instruct",
            model_config_dict={"temperature": 0},
        ),
        "image": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/Qwen2.5-VL-72B-Instruct",
            model_config_dict={"temperature": 0},
        ),
        "document": ModelFactory.create(
            model_platform=ModelPlatformType.SILICONFLOW,
            model_type="Qwen/Qwen2.5-VL-72B-Instruct",
            model_config_dict={"temperature": 0},
        ),
    }

    # Configure toolkits
    tools = [
        *BrowserToolkit(
            headless=False,  # Set to True for headless mode (e.g., on remote servers)
            web_agent_model=models["browsing"],
            planning_agent_model=models["planning"],
            output_language="Chinese",
        ).get_tools(),
        *VideoAnalysisToolkit(model=models["video"]).get_tools(),
        *CodeExecutionToolkit(sandbox="subprocess", verbose=True).get_tools(),
        *ImageAnalysisToolkit(model=models["image"]).get_tools(),
        SearchToolkit().search_duckduckgo,
        SearchToolkit().search_google,  # Comment this out if you don't have google search
        SearchToolkit().search_wiki,
        SearchToolkit().search_baidu,
        *ExcelToolkit().get_tools(),
        *DocumentProcessingToolkit(model=models["document"]).get_tools(),
        *FileWriteToolkit(output_dir="./").get_tools(),
    ]

    # Configure agent roles and parameters
    user_agent_kwargs = {"model": models["user"]}
    assistant_agent_kwargs = {"model": models["assistant"], "tools": tools}

    # Configure task parameters
    task_kwargs = {
        "task_prompt": question,
        "with_task_specify": False,
    }

    # Create and return the society
    society = RolePlaying(
        **task_kwargs,
        user_role_name="user",
        user_agent_kwargs=user_agent_kwargs,
        assistant_role_name="assistant",
        assistant_agent_kwargs=assistant_agent_kwargs,
    )

    return society


def main():
    r"""Main function to run the OWL system with an example question."""
    # Default research question
    default_task = "统计豆瓣电影top10的电影名称."

    # Override default task if command line argument is provided
    task = sys.argv[1] if len(sys.argv) > 1 else default_task

    # Construct and run the society
    society = construct_society(task)
    answer, chat_history, token_count = run_society(society)

    # Output the result
    print(f"\033[94mAnswer: {answer}\033[0m")


if __name__ == "__main__":
    main()

然后配置好下面的参数，具体地址都在括号里面。

# Google Search API (https://coda.io/@jon-dallas/google-image-search-pack-example/search-engine-id-and-google-api-key-3)
GOOGLE_API_KEY='AI********A'
SEARCH_ENGINE_ID='a********0'

# Chunkr API (https://chunkr.ai/)
CHUNKR_API_KEY='ch_********'

# Firecrawl API (https://www.firecrawl.dev/)
FIRECRAWL_API_KEY='fc-********'
#FIRECRAWL_API_URL="https://api.firecrawl.dev"

GoogleSearch

CHUNKR_API_KEY

FIRECRAWL_API_KEY

这些配置完成以后，就在终端运行该命令：python owl/webapp_zh.py ，启动web界面，并在浏览器中打开红框中的地址。

4. 测试 ​

还是问他老问题：统计豆瓣电影top10的电影名称 ，看看结果是否正确。

后端运行日志：

输出结果：

这个结果虽然也不完全正确，但是也还是比较接近了。

三、结语 ​

事情搞成这样，我也不想啊！本来我还打算多测一下的，但是实在是太晚了，身体要紧，其他的例子就留给大伙自由发挥吧。

关注我，了解更多AI黑科技

Copyright © 2019-present Ailot