Microsoft today introduced Autopilot as part of its new Copilot experience: a persistent, proactive personal agent that keeps working even when you’re not. Its foundation is OpenClaw.Microsoft 今日推出了作为其全新 Copilot 体验一部分的 Autopilot：这是一个持久、主动的个人代理，即使在你离开时也能持续工作。它的基础正是 OpenClaw。
Omar Shahine, who leads the team building it, put it directly:负责构建该产品的团队负责人 Omar Shahine 直言道：
“We are building Autopilot on @openclaw, working with @steipete and the OpenClaw Foundation to make it a fantastic enterprise grade runtime.”“我们正在基于 @openclaw 构建 Autopilot，并与 @steipete 及 OpenClaw 基金会合作，将其打造成一个出色的企业级运行时。”
Autopilot is the new name for Microsoft Scout, announced earlier this year. Microsoft’s launch post says Autopilot is expanding to private preview at the end of this month.Autopilot 是 Microsoft Scout 的新名称，该产品于今年早些时候宣布。Microsoft 的发布文章称，Autopilot 将于本月底扩展至私人预览版。
For the OpenClaw community, there is another part of the story worth highlighting: the improvements coming back upstream.对于 OpenClaw 社区而言，故事中还有另一个值得强调的部分：回馈到上游的改进。
Microsoft’s commitment is visible in both its public announcements and the work of contributors across the project. That work ranges from checking an agent’s security configuration to building a native Windows experience and making everyday agent interactions more dependable.Microsoft 的承诺不仅体现在其公开声明中，也体现在整个项目贡献者的工作中。这些工作涵盖了从检查代理的安全配置，到构建原生 Windows 体验，再到使日常代理交互更加可靠等各个方面。
Making an agent’s configuration checkable使代理的配置可检查化
An organization adopting a personal agent needs to answer practical questions. Which communication channels are enabled? Which model providers can it use? What can it access? Does its configuration still match the rules the organization intended?采用个人代理的组织需要回答一些实际问题。启用了哪些通信渠道？它可以使用哪些模型提供商？它可以访问什么？它的配置是否仍然符合组织预期的规则？
In its original Scout announcement, Microsoft made an explicit commitment:在最初的 Scout 发布公告中，Microsoft 做出了明确承诺：
“We are contributing policy conformance directly upstream to OpenClaw.”“我们正在直接向上游 OpenClaw 贡献策略合规性功能。”
Gio Della-Libera’s upstream work gives that commitment a concrete form. OpenClaw’s Policy plugin lets an operator describe requirements, compare them with the actual configuration, and produce a record of the result. Subsequent contributions extend those checks across model providers, networks and MCP servers, secret and authentication configuration, and other parts of an agent’s environment.Gio Della-Libera 的上游工作使这一承诺得以具体化。OpenClaw 的策略插件允许操作员描述需求，将其与实际配置进行比较，并生成结果记录。随后的贡献将这些检查扩展到了模型提供商、网络和 MCP 服务器、密钥与身份验证配置，以及代理环境的其他部分。
Omar also contributed message-routing checks, so operators can test whether representative incoming conversations reach the intended agent.Omar 还贡献了消息路由检查功能，以便操作员可以测试代表性的传入对话是否到达了预期的代理。
These checks let operators verify an agent’s configuration and detect when it no longer matches their requirements.这些检查让操作员能够验证代理的配置，并在其不再符合要求时及时发现。
A first-class home on Windows在 Windows 上的顶级体验
The Foundation and Microsoft’s Windows team have also described a long-term collaboration to make Windows a premier platform for OpenClaw.基金会和 Microsoft 的 Windows 团队还描述了一项长期合作计划，旨在使 Windows 成为 OpenClaw 的首选平台。
You can see that collaboration in the native Windows companion.你可以在原生 Windows 伴侣应用中看到这种合作。
Scott Hanselman’s contributions include a guided setup experience, alongside ongoing reliability and release work. Régis Brid contributed native WinUI chat and inline command approvals. Barbara Kudiess has worked across the native interface, model selection, approvals and sandbox integration. Caleb Eden’s contributions help the native chat display user and agent media properly.Scott Hanselman 的贡献包括引导式设置体验，以及持续的可靠性和发布工作。Régis Brid 贡献了原生 WinUI 聊天和内联命令批准功能。Barbara Kudiess 在原生界面、模型选择、批准和沙盒集成方面进行了大量工作。Caleb Eden 的贡献则帮助原生聊天界面正确显示用户和代理的媒体内容。
There is important work underneath that interface, too. Paul Campbell contributed the MXC sandbox backend for supported Windows environments. Alongside the companion’s own sandbox integration, it gives operators another way to constrain agent command execution using Microsoft’s execution-container technology.在该界面之下也有重要的工作。Paul Campbell 为受支持的 Windows 环境贡献了 MXC 沙盒后端。除了伴侣应用自身的沙盒集成外，它还为操作员提供了一种使用 Microsoft 执行容器技术来限制代理命令执行的新途径。
These practical improvements drive easier setup, clearer interaction, and more control over what an agent can do.这些实用的改进带来了更简单的设置、更清晰的交互，以及对代理行为更强的控制力。
Reliability for agents that keep working为持续运行的代理提供可靠性
An always-on agent has to handle more than a successful demo. It has to stay useful through background jobs, restarts, long conversations and failures.一个始终在线的代理不仅要处理成功的演示，还必须在后台作业、重启、长时间对话和故障中保持可用。
Galin Iliev’s contributions include prioritizing queued user requests over background work, fixing scheduler behavior that could hang the Gateway, and improving responsiveness during database recovery. His recent streamed-completion fix prevents the same Codex reply from appearing twice in conversation history when its streaming and completion identifiers differ.Galin Iliev 的贡献包括优先处理排队的用户请求而非后台工作，修复可能导致网关挂起的调度程序行为，以及提高数据库恢复期间的响应速度。他最近修复的流式完成问题，防止了当流式传输和完成标识符不同时，相同的 Codex 回复在对话历史中出现两次的情况。
Eduardo Piva contributed a guard against repeated tool loops after context compaction and work to reduce session-store memory retention.Eduardo Piva 贡献了一种针对上下文压缩后重复工具循环的防护措施，并致力于减少会话存储的内存占用。
Omar’s contributions include another subtle but important distinction: a command that definitely never ran is different from a command whose outcome is unknown. When an agent cannot tell whether an action executed, blindly repeating it can make the problem worse. The upstream change preserves that uncertainty and tells the agent not to rerun it automatically.Omar 的贡献还包括另一个微妙但重要的区别：明确未运行的命令与结果未知的命令是不同的。当代理无法判断操作是否执行时，盲目重复执行可能会使问题恶化。上游的这一更改保留了这种不确定性，并告知代理不要自动重新运行它。
These are the details that help turn an agent from something impressive into something people can depend on.正是这些细节将代理从令人印象深刻的工具转变为人们可以依赖的伙伴。
Better integrations, safer everyday interactions更好的集成，更安全的日常交互
Even the smallest contributions make noticeable improvements in daily use.即使是最小的贡献也能在日常使用中带来显著的改进。
Gio’s Microsoft Teams reaction fix lets an agent react to the message it is currently answering without having to copy an internal message ID. The change retains the boundary between the current conversation and another conversation.Gio 对 Microsoft Teams 反应功能的修复，让代理可以在无需复制内部消息 ID 的情况下，对当前正在回答的消息做出反应。此更改保留了当前对话与其他对话之间的界限。
Pengfei Ni contributed secret redaction in execution-approval prompts, helping keep recognizable API keys, tokens and passwords out of command text displayed in chat. Asking for permission should not unnecessarily expose a credential.Pengfei Ni 贡献了执行批准提示中的密钥脱敏功能，有助于防止可识别的 API 密钥、令牌和密码出现在聊天显示的命令文本中。请求权限时不应不必要地暴露凭据。
Kunal Karmakar improved Azure OpenAI provider support, wiring Azure endpoints to the appropriate Responses implementation so configured agents and tools can use them correctly.Kunal Karmakar 改进了 Azure OpenAI 提供商的支持，将 Azure 端点连接到相应的 Responses 实现，以便配置好的代理和工具能够正确使用它们。
Together with Eduardo’s memory-retention work and Galin’s duplicate-reply fix, these marginal improvements pack a mighty punch in the user experience.连同 Eduardo 的内存保留工作和 Galin 的重复回复修复，这些边际改进在用户体验上产生了巨大的影响。
Improvements that reach beyond one product超越单一产品的改进
One of the most human parts of this contribution story is Omar’s work on messaging. His upstream changes include iMessage approvals, native polls, and keeping replies in the right conversation.这一贡献故事中最具人文色彩的部分之一是 Omar 在消息传递方面的工作。他的上游更改包括 iMessage 批准、原生投票，以及确保回复保留在正确的对话中。
That work benefits OpenClaw users on Apple devices, too. It is a reminder that a shared project can improve across platforms even while companies build distinct products on top of it.这项工作也让 Apple 设备上的 OpenClaw 用户受益。这提醒我们，即使各家公司在项目之上构建了不同的产品，共享项目依然可以跨平台实现改进。
Not every contribution by someone at Microsoft is necessarily an Autopilot feature or a company-sponsored project. What the public record shows is a substantial body of work contributed back to OpenClaw, alongside Microsoft’s explicit commitment to upstream collaboration.并非 Microsoft 员工的每一项贡献都必然是 Autopilot 的功能或公司赞助的项目。公开记录显示的是大量回馈给 OpenClaw 的工作成果，以及 Microsoft 对上游协作的明确承诺。
Peter Steinberger welcomed that collaboration when it was announced in June:Peter Steinberger 在六月份宣布这一合作时表示欢迎：
Thank you to Omar, Gio, Scott, Barbara, Paul, Galin, Eduardo, Pengfei, Kunal, Régis, Caleb, and the other contributors and maintainers building this together.感谢 Omar、Gio、Scott、Barbara、Paul、Galin、Eduardo、Pengfei、Kunal、Régis、Caleb 以及其他共同构建此项目的贡献者和维护者。
OpenClaw remains an open infrastructure for building and customizing powerful agents, both for personal and organizational use. Microsoft’s Autopilot is a promising example of the possibilities, and we’re grateful to the contributors whose upstream work makes this platform more useful for everyone.OpenClaw 仍然是一个用于构建和定制强大代理的开放基础设施，既适用于个人也适用于组织。Microsoft 的 Autopilot 是一个充满希望的范例，我们感谢那些通过上游工作使该平台对所有人更加有用的贡献者。
Press Enter or Space to expand or collapse this image.按回车键或空格键展开或折叠此图像。