Memory on Claude Managed Agents is available today in public beta. Your agents can now learn from every session, using an intelligence-optimized memory layer that balances performance with flexibility. Because memories are stored as files, developers can export them, manage them via the API, and keep full control over what agents retain. Agents that learn across sessions Managed Agents pairs production infrastructure with a harness tuned for performance. Memory extends that: it’s optimized against internal benchmarks for long-running agents that improve across sessions and share what they've learned with each other. We've found that agents are most effective with memory when it builds on the tools they already use. Memory on Managed Agents mounts directly onto a filesystem, so Claude can rely on the same bash and code execution capabilities that make it effective at agentic tasks. With filesystem-based memory, our latest models save more comprehensive, well-organized memories and are more discerning about what to remember for a given task. Portable memories for production-grade agents Memory is built for enterprise deployments, with scoped permissions, audit logs, and full programmatic control. Stores can be shared across multiple agents with different access scopes. For example, an org-wide store might be read-only, while per-user stores allow reads and writes. Multiple agents can work concurrently against the same store without overwriting each other. Memories are files that can be exported and independently managed via the API, giving developers full control. All changes are tracked with a detailed audit log, so you can tell which agent and session a memory came from. You can roll back to an earlier version or redact content from history. Updates also surface in the Claude Console as session events, so developers can trace what an agent learned and where it came from. What teams are building Teams have been using memory to close feedback loops, speed up verification, and replace custom retrieval infrastructure: Netflix agents carry context across sessions, including insights that took multiple turns to uncover and corrections from a human mid-conversation, instead of manually updating prompts and skills. Rakuten's task-based long-running agents use memory to learn from every session and avoid repeating past mistakes, cutting first-pass errors by 97%, all within workspace-scoped, observable boundaries. Wisedocs built their document verification pipeline on Managed Agents, using cross-session memory to spot and remember recurring document issues, speeding up verification by 30%. Ando is building their workplace messaging platform on Managed Agents, capturing how each organization interacts instead of building memory infrastructure themselves.
Claude Managed Agents 的 Memory 现已以 public beta 形式提供。现在，你的 agent 可以从每一次 session 中学习，借助一层针对智能任务优化的 memory 层，在性能与灵活性之间取得平衡。由于 memory 以文件形式存储，开发者可以导出它们、通过 API 对其进行管理，并且对 agent 保留哪些内容保持完全控制。能够跨 session 学习的 agent：Managed Agents 将 production 基础设施与针对性能调优的 harness 相结合。Memory 在此基础上进一步扩展：它依据内部 benchmark 进行了优化，面向能够在长时间运行中跨 session 持续改进、并彼此共享所学内容的 agent。我们发现，当 memory 建立在 agent 已经使用的工具之上时，agent 的效果最好。Managed Agents 上的 Memory 直接挂载到文件系统，因此 Claude 可以依赖同样的 bash 和代码执行能力，而这些能力正是它在 agentic task 中高效表现的关键。借助基于文件系统的 memory，我们最新的模型能够保存更全面、组织更有条理的 memory，并且更善于判断针对特定任务应当记住什么。面向 production-grade agent 的可移植 memory：Memory 是为 enterprise 部署而构建的，具备 scoped permissions、audit logs 以及完整的 programmatic control。存储区可以在多个具有不同 access scope 的 agent 之间共享。例如，组织范围的存储区可以是只读的，而每用户存储区则允许读取和写入。多个 agent 可以同时在同一个存储区上工作，而不会相互覆盖。Memory 以文件形式存在，可以导出，也可以通过 API 独立管理，从而给予开发者完全控制。所有变更都会通过详细的 audit log 记录，因此你可以知道某条 memory 来自哪个 agent 和哪个 session。你还可以回滚到更早的版本，或从历史中 redact 内容。更新也会在 Claude Console 中显示为 session event，因此开发者可以追踪 agent 学到了什么，以及这些内容来自哪里。团队正在构建的内容：各团队一直在使用 memory 来闭合反馈回路、加快验证速度，并替代自定义的检索基础设施：Netflix 的 agent 能够跨 session 保留上下文，包括那些需要多轮对话才能发现的洞见，以及对话中人类提供的修正，而不再需要手动更新 prompt 和 skills。Rakuten 的基于任务的长时运行 agent 使用 memory 从每次 session 中学习，并避免重复过去的错误，在 workspace-scoped、可观测的边界内，将首次处理错误减少了 97%。Wisedocs 在 Managed Agents 之上构建了其文档验证流程，利用跨 session 的 memory 识别并记住反复出现的文档问题，将验证速度提升了 30%。Ando 正在基于 Managed Agents 构建其 workplace messaging platform，记录每个组织的交互方式，而不是自己搭建 memory 基础设施。