LangSmith

本文介绍了如何通过Auth Proxy来保护LangSmith代理沙箱的网络访问，确保在大规模部署代理时的安全性。Auth Proxy通过在网络层控制和管理代理与外部服务的交互，实现了凭据的安全管理、网络访问的显式控制以及团队职责的清晰分离。

轨迹标注需多维评估而非单一判断，Edith-1模型以1/274成本超越Sonnet-5.5，LangSmith框架实现难度与正确性分离评估。

LangChain团队通过成本可见性、控制和优化工具，连续两个月显著降低编码代理成本。

Clay通过LangSmith的insights产品和在线评估器实现百万级代理运行监控，揭示规模化部署中的行为分析挑战。

代理改进应聚焦于工具链优化而非模型本身，通过调整工具边界实现性能提升。

丰田通过LangChain和LangSmith实现制造AI，减少停机时间并获得六位数ROI。

LangSmith 现支持 AWS 自带云部署，数据保留在 AWS 边界内，LangChain 管理运维，兼顾数据主权与托管服务效率。

LangSmith沙箱为AI代理提供安全隔离的计算环境，支持秒级启动、代码执行和资源控制，已被7000+企业采用。

LLM评估框架存在可测量偏差，RAGAS/DeepEval/Promptfoo各有适用场景，需结合生产监控工具实现完整评估体系。

LangSmith 现已支持 Cursor、Copilot 等编码代理的标准化追踪，实现无需额外配置的全链路可观测性。

LangChain为Codex会话集成LangSmith追踪插件，实现工具调用、令牌使用等操作的实时追踪。

LangSmith新增对Cursor、Copilot等代码代理的全会话追踪功能，支持无需额外配置的稳定键标识与运行树分析。

施耐德电气通过自托管LangSmith平台管理全球60+ AI代理，覆盖100+国家，其AI助手服务16万员工，验证了LLMOps在大规模AI部署中的可行性。

GLM 5.2与dcode结合实现开源模型编码突破，性能接近闭源模型且成本更低。

LangSmith 通过将代理的交互痕迹转化为持久记忆，实现代理的持续学习与行为优化。

LangSmith 引入了‘睡眠时间计算’机制，通过后台分析代理轨迹并更新记忆存储，提升智能代理的长期记忆能力。

LangChain 推出 Engine，通过自动化分析代理日志和测试，提升代理开发效率。

可观测性对于智能代理是必不可少的，需结合 LangGraph 和 LangSmith 工具实现。

LangSmith Sandboxes securely run untrusted agent code via isolated execution environments, effectively preventing risks like the 'sci-holude' supply chain attack, applicable in AI agent scenarios for software engineering and data analysis.

LangChain introduces the Agent Development Lifecycle (ADLC), dividing agent development into four phases—build, test, deploy, and monitor—emphasizing that its fundamental difference from traditional software lies in infinite input/output spaces and non-determinism, with successful teams adopting a "ship early, iterate fast" pattern.

This article introduces seven leading LLM observability tools that help AI engineers monitor, evaluate, and debug large language model applications running in production.

Using a sandbox environment for AI agents significantly reduces production risks, especially during high-risk operations by isolating errors and preventing data corruption.

LangSmith’s Context Hub provides a centralized, versioned context management solution to address AI Agent failures caused by missing, outdated, or fragmented context; it enables human-editable and agent-readable collaboration via agent.md contracts and memory folders.

LangChain has released Mission Control, a decoupled, in-cluster application for deploying, configuring, observing, and troubleshooting self-hosted LangSmith and related LangChain infrastructure.

Top AI agent builders consistently exhibit two traits: systemic thinking and iterative improvement, enabling robust task decomposition and data-driven optimization.

LangSmith平台提供完整的代理开发生命周期管理工具，支持构建、测试、部署和监控改进型代理。

Google ADK agents can now be directly deployed to LangSmith Deployments, with a wrapper package enabling production-ready persistence, streaming, and tracing for agents built with Google's Agent Development Kit.

LangSmith Sandboxes GA release introduces snapshots and cheap forks, enabling efficient parallel development and error recovery.

In financial services, explaining how a conclusion was reached matters as much as the conclusion itself; LangChain uses LangSmith to log every query, response, and intermediate result, enabling full traceability of AI agent decisions for transparency and compliance.

AI still needs better data for more sophisticated answers, especially for finance agents. LangChain's team achieved significant performance gains using Deep Agents, LangSmith, and You.com Finance API, showing rare improvements in competitive benchmarks where multiple competitors also published results.