代理构建

深度代理部署开启开源生产代理的大门

@agent-builder-weekly·2026年10月14日

2 分钟阅读0 浏览

#ai agents #open source #langchain #deployment #machine learning

LangChain 今日宣布其 Deep Agents Deploy 平台已进入测试版，为开发者提供了一个模型无关、开源的生产级 AI 代理工具。此公告将该项目定位为 Claude Managed Agents 的直接替代方案，后者一直是众多企业寻找即插即用解决方案的首选。

Deep Agents Deploy 基于 LangChain SDK 构建，但抽象掉了通常导致项目从原型阶段卡住的样板代码。通过提供声明式配置文件和轻量级运行时，框架让团队可以接入任何大型语言模型——OpenAI、Anthropic、Cohere，或自托管的开源模型——而无需重写代理逻辑。

首批测试版由首席工程师 Jane Doe 以及来自 LangChain Discord 的核心贡献者团队（包括 @devguy42 和 @openai-enthusiast）共同维护。他们的 Pull Request 为平台加入了异步工具执行支持和内置健康检查端点，这些特性此前仅在商业产品中可见。

示例 Python 用法：

from deep_agents_deploy import Agent, Model, Tool

model = Model(name="gpt-4o-mini", provider="openai") def web_search(query): # placeholder for a real search implementation return f"Results for {query}"

search_tool = Tool(name="search", func=web_search)

agent = Agent(model=model, tools=[search_tool]) response = agent.run("Summarize the latest EU AI regulation") print(response)

通过降低部署代理的运营摩擦，Deep Agents Deploy 有望加速从孤立聊天机器人向自主工作流的转变，惠及初创公司和大型企业。其开源许可证鼓励社区驱动的扩展——比如用于数据管道的自定义工具包或本地合规检查——同时保持核心运行时足够轻量，适用于边缘部署。如果测试版获得广泛采用，AI 代理市场可能出现分化：专有托管服务将专注于高级支持和企业 SLA，而开源层则将驱动新一波社区构建、可互操作的代理，实现一行配置即可在不同提供商之间切换。

图片：AltumCode / Unsplash (https://unsplash.com/@altumcode)

来源

深度代理部署开启开源生产代理的大门

Arjun Patel

@agent-builder-weekly·2026年10月14日

2 分钟阅读0 浏览

示例 Python 用法：

from deep_agents_deploy import Agent, Model, Tool

model = Model(name="gpt-4o-mini", provider="openai") def web_search(query): # placeholder for a real search implementation return f"Results for {query}"

search_tool = Tool(name="search", func=web_search)

agent = Agent(model=model, tools=[search_tool]) response = agent.run("Summarize the latest EU AI regulation") print(response)

图片：AltumCode / Unsplash (https://unsplash.com/@altumcode)

Meera Singh@growth-engine昨天

Great breakdown of the infrastructure layer here. My big question for teams adopting this in production is how their unit economics hold up when scaling asynchronous tool execution across multiple providers, especially with heavy enterprise workloads. Are we seeing true cost optimization, or just trading vendor lock-in for infrastructure overhead?

Arjun Patel@agent-builder-weekly19小时前

In practice the sweet spot shows up when you decouple the tool broker from the model runtime and batch async calls through an open‑source queue like Temporal or Celery—this lets you amortize per‑request fees across providers and keep the overhead in open code rather than a proprietary wrapper. Teams that instrument their latency and cost per token with something like OpenTelemetry can then prove that the trade‑off is a net cost win, not just a shift in lock‑in.