OpenClaw 与 LangChain:我们为何不使用框架

📄 中文摘要

OpenClaw 是一个 npm 包,安装后作为本地进程运行,为 Claude 提供访问工具的能力,包括文件系统、shell、MCP 服务器和内存。它是一个运行时,而非框架,不会规定代理逻辑的组织方式。OpenClaw 对工具调用的方式有明确的看法,但对代理的具体功能没有限制。这一差异使得 OpenClaw 更加灵活,能够满足特定需求,而不受框架的约束。与之相比,LangChain 作为主流的 Python 框架,虽然拥有庞大的生态系统和处理许多底层工作的能力,但可能不适合所有场景。

📄 English Summary

OpenClaw vs LangChain: Why We Don't Use Frameworks

OpenClaw is an npm package that runs as a local process, providing Claude with access to various tools such as the file system, shell, MCP servers, and memory. It functions as a runtime rather than a framework, meaning it does not dictate how to organize agent logic. OpenClaw has specific opinions on how tools are called but does not impose restrictions on what the agent can do. This distinction allows OpenClaw to be more flexible and tailored to specific needs without the constraints of a framework. In contrast, LangChain, as the dominant Python framework, offers a vast ecosystem and handles much of the plumbing, but may not be suitable for every scenario.

Powered by Cloudflare Workers + Payload CMS + Claude 3.5

数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等