初嘗 OpenManus

記錄一下用 uv 安裝 OpenManus

長期沒有更新這個 Blog 呢…

最近有個叫 Manus 的 AI agent, 類似 AutoGPT 那種。

然後很快 MetaGPT 團隊的人很快弄了個開源版。

趁周末便嘗試安裝一下,看看是否實用,有沒有什麼特別。

如果好用,我就可能不打算自建 n8n 或 kestra 了。

uv 安裝

1
2
git clone https://github.com/mannaandpoem/OpenManus.git
cd OpenManus

因為我目前在用 uv 去管理 Python

建立 virtual environment 和安裝 packages

1
2
uv venv
uv pip install -r requirements.txt

Playwright 安裝 Chromium

1
uv run playwright install

設置

更改 API 設置

1
cp config/config.example.toml config/config.toml

我使用 DeepSeek v3 API,於是便將 config/config.toml 刪改成:

1
2
3
4
5
6
7
# Global LLM configuration
[llm]
model = "deepseek-chat"
base_url = "https://api.deepseek.com"
api_key = "<YOUR_API>"
# max_tokens = 4096
temperature = 0.0

其他

我見只能輸入一行,不太好用。便把 main.py 中的

1
            prompt = input("Enter your prompt (or 'exit' to quit): ")

改成了

1
2
3
4
5
6
7
8
            print("Enter your prompt (or 'exit' to quit). End input with an empty line:")
            lines = []
            while True:
                line = input()
                if line == "":
                    break
                lines.append(line)
            prompt = "\n".join(lines)

運行

1
uv run main.py

用後感

只用了一塊錢的 API 成本玩了會。覺得效果不怎樣。各方面都有提升空間。

譬如我這可以改用其他 model 的 API (例如支持 function calling 的 model), 修改 Prompts 之類, 加入 vision 用的 model。

Manus 官網 demo 視頻的效果遠比這個開源版的好太多。

也有點想試 Opera 的 Browser Operator

References

https://github.com/mannaandpoem/OpenManus

https://manus.im/

https://press.opera.com/2025/03/03/opera-browser-operator-ai-agentics/

comments powered by Disqus
Built with Hugo
Theme Stack designed by Jimmy