{
  "schema_version": 1,
  "report_date": "2026-06-01",
  "title": "AI 日报 2026-06-01",
  "summary": "今天主体信息扩展到 10 条：计费与成本归因、企业 agent 平台、国产模型、physical AI、企业 BI、AIGC 内容工具和科研应用一起进入日报视野。模型可用性 incident 被放入轻量运营线索，模型发布只保留 MiniMax M3 和 Cosmos 3 这类首发。",
  "hero_highlights": [
    {
      "title": "Copilot usage-based billing 生效",
      "url": "https://github.blog/changelog/2026-04-27-github-copilot-code-review-will-start-consuming-github-actions-minutes-on-june-1-2026/",
      "reason": "AI 开发工具成本归因从 seat 走向用量、review 和 Actions minutes 联动。"
    },
    {
      "title": "MiniMax M3 与 NVIDIA Cosmos 3 同日进入模型发布",
      "url": "https://www.minimax.io/blog/minimax-m3",
      "reason": "国产 coding/多模态与 physical AI open model 同时值得跟踪。"
    },
    {
      "title": "Vercel AI Gateway spend caps 进入常规运营观察",
      "url": "https://x.com/rauchg/status/2060787704166776927",
      "reason": "AI 网关的 API key 级预算控制会影响团队成本治理。"
    }
  ],
  "canonical_url": "https://jasonxzwen.github.io/ai-daily-cn/reports/2026/06/2026-06-01.html",
  "html_path": "reports/2026/06/2026-06-01.html",
  "github_trending": [
    {
      "repo": "harry0703/MoneyPrinterTurbo",
      "name": "harry0703/MoneyPrinterTurbo",
      "description": "一键生成短视频的 AIGC 工具，覆盖文案、素材、配音和视频组装流程。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/harry0703/MoneyPrinterTurbo",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 1,
      "previous_rank": 2,
      "rank_delta": 1,
      "trend": "up",
      "importance": "notable"
    },
    {
      "repo": "microsoft/markitdown",
      "name": "microsoft/markitdown",
      "description": "Microsoft 开源的文档转 Markdown 工具，可把多种文件转成适合 LLM/RAG 管线使用的文本。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/microsoft/markitdown",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 2,
      "previous_rank": 1,
      "rank_delta": -1,
      "trend": "down",
      "importance": "general"
    },
    {
      "repo": "D4Vinci/Scrapling",
      "name": "D4Vinci/Scrapling",
      "description": "自适应网页抓取框架，覆盖动态页面、crawl 和结构化数据提取。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/D4Vinci/Scrapling",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 3,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "notable"
    },
    {
      "repo": "nesquena/hermes-webui",
      "name": "nesquena/hermes-webui",
      "description": "Hermes Agent 的 Web 和移动端界面，关注 agent 任务跨设备访问。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/nesquena/hermes-webui",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 4,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "EveryInc/compound-engineering-plugin",
      "name": "EveryInc/compound-engineering-plugin",
      "description": "面向 Claude Code、Codex、Cursor 的工程插件，连接 coding agent 与团队流程。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/EveryInc/compound-engineering-plugin",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 5,
      "previous_rank": 1,
      "rank_delta": -4,
      "trend": "down",
      "importance": "general"
    },
    {
      "repo": "github/docs",
      "name": "github/docs",
      "description": "docs.github.com 开源仓库，用于跟踪 GitHub 产品文档和行为说明。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/github/docs",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 6,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "OpenBMB/VoxCPM",
      "name": "OpenBMB/VoxCPM",
      "description": "面向声音克隆、创意音色和多语音生成的 tokenizer-free TTS 项目。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/OpenBMB/VoxCPM",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 7,
      "previous_rank": 4,
      "rank_delta": -3,
      "trend": "down",
      "importance": "general"
    },
    {
      "repo": "revfactory/harness",
      "name": "revfactory/harness",
      "description": "生成领域专用 agent team、agent 和 skills 的 meta-skill 项目。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/revfactory/harness",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 8,
      "previous_rank": 10,
      "rank_delta": 2,
      "trend": "up",
      "importance": "general"
    },
    {
      "repo": "FareedKhan-dev/train-llm-from-scratch",
      "name": "FareedKhan-dev/train-llm-from-scratch",
      "description": "从数据下载到文本生成的 LLM 训练教程型仓库。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/FareedKhan-dev/train-llm-from-scratch",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 9,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "supermemoryai/supermemory",
      "name": "supermemoryai/supermemory",
      "description": "面向 AI 应用的 memory engine 和可扩展 memory API。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/supermemoryai/supermemory",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "language": "all",
      "window": "daily",
      "rank": 10,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    }
  ],
  "main_items": [
    {
      "title": "GitHub Copilot 计费切换今天生效，代码审查进入成本归因",
      "event_date": "2026-06-01",
      "url": "https://github.blog/changelog/2026-04-27-github-copilot-code-review-will-start-consuming-github-actions-minutes-on-june-1-2026/",
      "source": "GitHub Changelog",
      "tier": "T0",
      "entities": [
        "GitHub Copilot",
        "AI Credits",
        "Actions minutes",
        "code review"
      ],
      "summary": "今天生效：GitHub Copilot code review 从 2026-06-01 起在私有仓库消耗 Actions minutes，公有仓库 Actions minutes 仍免费。 双计费：每次 review 同时进入 Copilot AI Credits 和 GitHub Actions minutes，覆盖 Pro、Pro+、Business、Enterprise。 管理员需要核对 Actions entitlement / budgets / runner 设置，否则自动 PR review 会变成新的 CI 成本项。",
      "bullets": [
        "**今天生效**：GitHub Copilot code review 从 2026-06-01 起在私有仓库消耗 Actions minutes，公有仓库 Actions minutes 仍免费。",
        "**双计费**：每次 review 同时进入 Copilot AI Credits 和 GitHub Actions minutes，覆盖 Pro、Pro+、Business、Enterprise。",
        "==管理员需要核对 **Actions entitlement / budgets / runner 设置**，否则自动 PR review 会变成新的 CI 成本项。=="
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "Microsoft Foundry 更新 agent 评测、成本归因和本地运行能力",
      "event_date": "2026-05-30",
      "url": "https://devblogs.microsoft.com/foundry/whats-new-in-microsoft-foundry-may-2026/",
      "source": "Microsoft Foundry Blog",
      "tier": "T0",
      "entities": [
        "Microsoft Foundry",
        "trace evaluation",
        "Foundry Local",
        "cost attribution"
      ],
      "summary": "评测与追踪：Foundry 把托管/外部 agent traces、评测和项目视角成本归因放在同一组更新里。 本地能力：Foundry Local 1.1 带来实时 ASR、embedding、Qwen 3.5 Vision、WebGPU 插件；1.2 又补多语 ASR、ARM64、WinML 2.0。 成本归因只解释模型/项目使用，完整账单仍要结合 Azure Cost Management、Search、Storage、Key Vault 等资源看。",
      "bullets": [
        "**评测与追踪**：Foundry 把托管/外部 agent traces、评测和项目视角成本归因放在同一组更新里。",
        "**本地能力**：Foundry Local 1.1 带来实时 ASR、embedding、Qwen 3.5 Vision、WebGPU 插件；1.2 又补多语 ASR、ARM64、WinML 2.0。",
        "==成本归因只解释模型/项目使用，完整账单仍要结合 Azure Cost Management、Search、Storage、Key Vault 等资源看。=="
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "MiniMax M3 发布，主打 coding、长上下文和原生多模态",
      "event_date": "2026-06-01",
      "url": "https://www.minimax.io/models/text/m3",
      "source": "MiniMax model page",
      "tier": "T0",
      "entities": [
        "MiniMax M3",
        "coding model",
        "1M context",
        "multimodal"
      ],
      "summary": "三能力合一：M3 主打 open-weight、coding/agentic、1M context 和原生多模态，API 最低保证 512K context。 长任务数据：官方案例给出 ICLR 论文复现 12 小时、18 commits、23 figures，以及 FP8 GEMM 147 次提交、9.4x speedup。 开发者侧有 Token Plan、API、MiniMax Code 和待开源本地部署路径，价格/配额会直接影响 coding agent 使用成本。",
      "bullets": [
        "**三能力合一**：M3 主打 open-weight、coding/agentic、1M context 和原生多模态，API 最低保证 512K context。",
        "**长任务数据**：官方案例给出 ICLR 论文复现 12 小时、18 commits、23 figures，以及 FP8 GEMM 147 次提交、9.4x speedup。",
        "开发者侧有 Token Plan、API、MiniMax Code 和待开源本地部署路径，==价格/配额会直接影响 coding agent 使用成本==。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "NVIDIA Cosmos 3 面向 physical AI 推理、世界模型和动作模型",
      "event_date": "2026-06-01",
      "url": "https://developer.nvidia.com/blog/develop-physical-ai-reasoning-world-and-action-models-with-nvidia-cosmos-3/",
      "source": "NVIDIA Developer Blog",
      "tier": "T0",
      "entities": [
        "NVIDIA Cosmos 3",
        "physical AI",
        "robotics",
        "world model"
      ],
      "summary": "架构变化：Cosmos 3 用 MoT 把 reasoner tower 和 generator tower 合并，输入可含文本、图像、视频、音频、动作。 模型规格：Nano 为 8B，面向工作站实时推理；Super 为 32B，面向 Hopper/Blackwell 数据中心部署。 NVIDIA 同步开放 6 类 synthetic datasets，并提供 BF16、FP8、NVFP4 NIM 路径，物理 AI 不只是视频生成。",
      "bullets": [
        "**架构变化**：Cosmos 3 用 MoT 把 reasoner tower 和 generator tower 合并，输入可含文本、图像、视频、音频、动作。",
        "**模型规格**：Nano 为 8B，面向工作站实时推理；Super 为 32B，面向 Hopper/Blackwell 数据中心部署。",
        "NVIDIA 同步开放 6 类 synthetic datasets，并提供 BF16、FP8、NVFP4 NIM 路径，==物理 AI 不只是视频生成==。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "NVIDIA 用 DOCA In-Silicon Security 强化 agentic AI 基础设施",
      "event_date": "2026-06-01",
      "url": "https://developer.nvidia.com/blog/advancing-ai-infrastructure-for-agentic-ai-with-nvidia-doca-in-silicon-security/",
      "source": "NVIDIA Developer Blog",
      "tier": "T0",
      "entities": [
        "NVIDIA DOCA",
        "In-Silicon Security",
        "AI factory",
        "agentic AI"
      ],
      "summary": "安全边界下沉：BlueField DPU 把监控、策略执行、遥测放到独立信任域，主机被攻破时仍可执行控制。 性能指标：DOCA 文章给出 runtime threat detection 最高 1,000x、网络/文件访问策略执行最高 800 Gb/s。 DOCA Argus、Vault、Flow 分别对应运行时威胁检测、文件级 zero-trust 访问和硬件加速网络策略。",
      "bullets": [
        "**安全边界下沉**：BlueField DPU 把监控、策略执行、遥测放到独立信任域，主机被攻破时仍可执行控制。",
        "**性能指标**：DOCA 文章给出 runtime threat detection 最高 1,000x、网络/文件访问策略执行最高 800 Gb/s。",
        "==DOCA Argus、Vault、Flow 分别对应运行时威胁检测、文件级 zero-trust 访问和硬件加速网络策略。=="
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "NVIDIA Alpamayo 文章聚焦自动驾驶模型闭环后训练",
      "event_date": "2026-06-01",
      "url": "https://developer.nvidia.com/blog/how-to-post-train-autonomous-vehicle-models-in-closed-loop-with-nvidia-alpamayo/",
      "source": "NVIDIA Developer Blog",
      "tier": "T0",
      "entities": [
        "NVIDIA Alpamayo",
        "autonomous vehicle",
        "closed-loop post-training"
      ],
      "summary": "训练范式：AlpaGym 把 AlpaSim closed-loop rollouts 接入策略训练，让模型从自己动作造成的后果中学习。 工程依赖：教程要求 CUDA/cuDNN、NCCL、Redis、Git LFS、Hugging Face auth，并用 Hydra 配置 policy、scene、reward。 输出关注 mean reward、failure rates、policy loss、rollout throughput 和 checkpoint，可用于 AV 模型闭环验收。",
      "bullets": [
        "**训练范式**：AlpaGym 把 AlpaSim closed-loop rollouts 接入策略训练，让模型从自己动作造成的后果中学习。",
        "**工程依赖**：教程要求 CUDA/cuDNN、NCCL、Redis、Git LFS、Hugging Face auth，并用 Hydra 配置 policy、scene、reward。",
        "==输出关注 mean reward、failure rates、policy loss、rollout throughput 和 checkpoint，可用于 AV 模型闭环验收。=="
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "阿里云 Quick BI 推 ticket-based enhanced embedding 处理数据分享权限",
      "event_date": "2026-06-01",
      "url": "https://www.alibabacloud.com/blog/sharing-data-without-risking-leaks-let-ticket-based-enhanced-embedding-strike-the-perfect-balance_603197",
      "source": "Alibaba Cloud Blog",
      "tier": "T0",
      "entities": [
        "Alibaba Cloud",
        "Quick BI",
        "ticket-based embedding",
        "data sharing"
      ],
      "summary": "权限模型：Quick BI ticket-based enhanced embedding 支持时间锁、访问次数锁和按角色个性化视图。 嵌入粒度：可嵌完整 dashboard、单个可视化组件、ad hoc query 和自助分析模块到 ERP、OA、CRM、App 或小程序。 治理看板跟踪授权用户、嵌入报表数、访问量和启用报表数，BI 分享从链接分发变成可运营服务。",
      "bullets": [
        "**权限模型**：Quick BI ticket-based enhanced embedding 支持时间锁、访问次数锁和按角色个性化视图。",
        "**嵌入粒度**：可嵌完整 dashboard、单个可视化组件、ad hoc query 和自助分析模块到 ERP、OA、CRM、App 或小程序。",
        "治理看板跟踪授权用户、嵌入报表数、访问量和启用报表数，==BI 分享从链接分发变成可运营服务==。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "Lingyang 在 Qwen Conference 展示 Quick BI 企业级数据方案",
      "event_date": "2026-06-01",
      "url": "https://www.alibabacloud.com/blog/lingyang-debuts-at-the-qwen-conference-in-singapore-quick-bi-deconstructs-enterprise-grade-data-solutions-for-the-ai-era_603196",
      "source": "Alibaba Cloud Blog",
      "tier": "T0",
      "entities": [
        "Lingyang",
        "Qwen Conference",
        "Quick BI",
        "enterprise data"
      ],
      "summary": "会议场景：Qwen Conference Singapore 上，Quick BI 把 enterprise Agentic Analytics 讲成“Goals -> Inferences -> Actions”。 业务覆盖：文章列出电商运营、销售运营、制造运营、供应链、金融运营等场景，强调统一指标和权限边界。 案例称手工日报工作量下降约 90%、问题发现解决提速 10x，报表迁移人工校验工作量下降 50% 以上。",
      "bullets": [
        "**会议场景**：Qwen Conference Singapore 上，Quick BI 把 enterprise Agentic Analytics 讲成“Goals -> Inferences -> Actions”。",
        "**业务覆盖**：文章列出电商运营、销售运营、制造运营、供应链、金融运营等场景，强调统一指标和权限边界。",
        "==案例称手工日报工作量下降约 90%、问题发现解决提速 10x，报表迁移人工校验工作量下降 50% 以上。=="
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    }
  ],
  "model_releases": [],
  "hot_blogs": [
    {
      "title": "Comprehensive observability for Amazon SageMaker AI LLM inference",
      "url": "https://aws.amazon.com/blogs/machine-learning/comprehensive-observability-for-amazon-sagemaker-ai-llm-inference-from-gpu-utilization-to-llm-quality/",
      "publisher": "AWS Machine Learning Blog",
      "author": "Sandeep Raveesh-Babu and Jonathan Kola",
      "event_date": "2026-05-29",
      "topic": "LLM inference observability and quality monitoring",
      "summary": "AWS 这篇文章把 LLM 推理可观测性拆成两个维度：服务基础设施的 quantity 指标和输出质量的 quality 指标。前者覆盖 invocation、latency、error、GPU/CPU 使用率和推理组件维度，后者通过采样与评测捕捉模型漂移、回答不一致或安全问题。它给出的架构使用 SageMaker AI Inference Components、CloudWatch 和 Amazon Managed Grafana，把不同 LLM 或不同 inference component 放在同一 endpoint 下观察。对生产团队来说，重点是不要只看 GPU 和延迟；LLM endpoint 可以运行健康但回答质量变差，也可能回答质量可接受但资源过度配置。",
      "content_type": "engineering_note",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "title": "Building Agentic Enterprises on AWS with the AWS for SAP MCP Server",
      "url": "https://aws.amazon.com/blogs/awsforsap/building-agentic-enterprises-on-aws-using-aws-for-sap-mcp-server-on-amazon-bedrock-agentcore/",
      "publisher": "AWS for SAP Blog",
      "author": "Rengarajan Sridharan and Krishnakumar Ramadoss",
      "event_date": "2026-05-29",
      "topic": "MCP server, SAP business data and Bedrock AgentCore",
      "summary": "AWS 说明 AWS for SAP MCP Server 如何运行在 Amazon Bedrock AgentCore Runtime 上，把 SAP OData API 暴露为 MCP tools，让 MCP client 和 agent 访问财务、采购、物流等业务流程。文章强调解耦 agent 与工具，使用 MCP 连接外部数据和工具，同时用 A2A 支持 agent 间协作。部署侧，它把 MCP server 作为容器镜像运行在客户 VPC 中，并结合 Bedrock AgentCore Identity、私有连接和会话隔离处理企业安全边界。对大型企业来说，这类方案的价值在于把 agent 从 demo 接入推进到真实 ERP 数据和流程，但也要求团队先整理 SAP API、权限、网络路径和审计责任。",
      "content_type": "engineering_note",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "title": "Data Formulator 0.7: AI-powered data analytics for enterprise data",
      "url": "https://www.microsoft.com/en-us/research/blog/data-formulator-0-7-ai-powered-data-analytics-for-enterprise-data/",
      "publisher": "Microsoft Research Blog",
      "author": "Chenglong Wang, Scott Tsukamaki, Michel Galley, Jianfeng Gao",
      "event_date": "2026-05-28",
      "topic": "AI agents for enterprise data analytics",
      "summary": "Microsoft Research 发布 Data Formulator 0.7，把企业数据连接、agent-guided exploration 和可视化迭代放进一个共享工作区。Data Connectors 支持数据库、数仓、BI 系统、对象存储和本地文件的持久连接、认证、预览和 metadata；context-aware agents 可以查看分析工作区、已加载表、既有图表和用户目标，再通过工具准备数据、写代码、生成 chart spec 并展示中间步骤。文章值得关注的是交互形态：它不是单轮聊天生成图表，而是把长分析会话、分支探索、图表直接编辑和可复现代码串在一起，适合企业数据团队评估 AI analytics 是否能进入 governed workflow。",
      "content_type": "blog",
      "image_urls": [],
      "importance": "notable"
    }
  ],
  "projects": [
    {
      "name": "harry0703/MoneyPrinterTurbo",
      "description": "一键生成短视频的 AIGC 工具，今日 GitHub Trending daily rank 1，适合观察内容生产自动化从脚本走向产品化模板。",
      "domains": [
        "AIGC_video",
        "content_automation",
        "short_video"
      ],
      "use_case": "用于把主题、文案、配音和素材组装成短视频工作流，适合内容团队评估自动化生产链。",
      "url": "https://github.com/harry0703/MoneyPrinterTurbo",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "D4Vinci/Scrapling",
      "description": "自适应网页抓取框架，今日 GitHub Trending daily rank 3，覆盖动态页面和大规模 crawl 场景。",
      "domains": [
        "web_scraping",
        "data_ingestion",
        "RAG"
      ],
      "use_case": "用于给 RAG、竞品监控或评测样本构建网页采集层。",
      "url": "https://github.com/D4Vinci/Scrapling",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "EveryInc/compound-engineering-plugin",
      "description": "面向 Claude Code、Codex、Cursor 等工具的工程插件，关注多工具协作、流程约束和代码生成质量。",
      "domains": [
        "coding_agent",
        "developer_workflow",
        "automation"
      ],
      "use_case": "用于评估团队如何把多个 coding agent 接入同一工程工作流。",
      "url": "https://github.com/EveryInc/compound-engineering-plugin",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "OpenBMB/VoxCPM",
      "description": "面向多语音生成、创意音色设计和声音克隆的 tokenizer-free TTS 项目，继续在 daily 榜前列。",
      "domains": [
        "AIGC_audio",
        "TTS",
        "voice_clone"
      ],
      "use_case": "用于播客、短视频和多语种内容生产中的声音生成与克隆评估。",
      "url": "https://github.com/OpenBMB/VoxCPM",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "supermemoryai/supermemory",
      "description": "面向 AI 应用的 memory engine 和 app，今日 GitHub Trending daily rank 10，定位为高速可扩展的 AI memory API。",
      "domains": [
        "agent_memory",
        "knowledge_base",
        "developer_api"
      ],
      "use_case": "用于给 agent 或个人 AI 应用提供跨会话记忆、检索和上下文管理能力。",
      "url": "https://github.com/supermemoryai/supermemory",
      "event_date": "2026-06-01",
      "source": "GitHub Trending daily",
      "signal": "trending",
      "importance": "notable"
    }
  ],
  "builder_observations": [
    {
      "author": "Guillermo Rauch",
      "content": "Rauch 提到 Vercel AI Gateway 的 per-API key spend caps。对 builder 来说，模型路由之外的关键变化是按项目、环境或客户拆分预算，减少共享 key 带来的成本不可解释问题。",
      "url": "https://x.com/rauchg/status/2060787704166776927",
      "role": "Vercel CEO",
      "event_date": "2026-05-30",
      "source": "follow-builders X feed",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "author": "Ryo Lu",
      "content": "Ryo Lu 说 Cursor auto-review 会解释命令和风险，让新开发者更容易判断下一步。这里的观察点是 agent 工具正在把风险提示、命令解释和执行确认做成默认交互。",
      "url": "https://x.com/ryolu_/status/2060766674203353190",
      "role": "Cursor product/design builder",
      "event_date": "2026-05-30",
      "source": "follow-builders X feed",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "author": "Peter Steinberger",
      "content": "Steinberger 说 GPT-5.5、/goal、autoreview 和 crabbox 让自己的 prompts 从约 30-60 分钟延长到 4-10 小时任务。长任务的验收、回滚和成本上限会成为 coding agent 日常管理项。",
      "url": "https://x.com/steipete/status/2060678430031597696",
      "role": "Builder / engineer",
      "event_date": "2026-05-30",
      "source": "follow-builders X feed",
      "image_urls": [],
      "importance": "notable"
    }
  ],
  "evidence_assets": [
    {
      "type": "figure",
      "title": "MiniMax M3 论文复现实验截图",
      "source_url": "https://www.minimax.io/models/text/m3",
      "local_path": "assets/evidence/minimax-m3-paper-reproduction.png",
      "caption": "图 1：MiniMax M3 官方页面展示的 ICLR 论文复现实验结果截图，保留原文图片而不是转写成表格。",
      "extraction_status": "source_image"
    },
    {
      "type": "figure",
      "title": "Cosmos 3 架构图",
      "source_url": "https://developer.nvidia.com/blog/develop-physical-ai-reasoning-world-and-action-models-with-nvidia-cosmos-3/",
      "local_path": "assets/evidence/nvidia-cosmos-3-architecture.webp",
      "caption": "图 2：NVIDIA Developer Blog 原文中的 Cosmos 3 reasoner/generator 架构图。",
      "extraction_status": "source_image"
    },
    {
      "type": "figure",
      "title": "DOCA Argus 威胁检测架构图",
      "source_url": "https://developer.nvidia.com/blog/advancing-ai-infrastructure-for-agentic-ai-with-nvidia-doca-in-silicon-security/",
      "local_path": "assets/evidence/nvidia-doca-argus-architecture.webp",
      "caption": "图 3：NVIDIA Developer Blog 原文中的 DOCA Argus AI threat detection 架构图。",
      "extraction_status": "source_image"
    },
    {
      "type": "figure",
      "title": "SageMaker LLM 推理可观测性架构图",
      "source_url": "https://aws.amazon.com/blogs/machine-learning/comprehensive-observability-for-amazon-sagemaker-ai-llm-inference-from-gpu-utilization-to-llm-quality/",
      "local_path": "assets/evidence/aws-sagemaker-llm-observability-architecture.png",
      "caption": "AWS Machine Learning Blog 原文首图：SageMaker endpoint、Inference Components、CloudWatch 与 Grafana 的监控链路。",
      "extraction_status": "source_image"
    },
    {
      "type": "figure",
      "title": "AWS for SAP MCP Server 架构图",
      "source_url": "https://aws.amazon.com/blogs/awsforsap/building-agentic-enterprises-on-aws-using-aws-for-sap-mcp-server-on-amazon-bedrock-agentcore/",
      "local_path": "assets/evidence/aws-sap-mcp-agentcore-architecture.png",
      "caption": "AWS for SAP Blog 原文架构图：MCP Server 运行在 Bedrock AgentCore Runtime 并连接 SAP OData API。",
      "extraction_status": "source_image"
    }
  ],
  "generated_at": "2026-06-01T16:00:00+08:00",
  "stories": [
    {
      "story_id": "main-github-copilot-billing-june1",
      "title": "GitHub Copilot 计费切换今天生效，代码审查进入成本归因",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "GitHub Changelog",
      "event_type": "update",
      "object": "GitHub Copilot 计费切换今天生效，代码审查进入成本归因",
      "what_happened": "今天生效：GitHub Copilot code review 从 2026-06-01 起在私有仓库消耗 Actions minutes，公有仓库 Actions minutes 仍免费。 双计费：每次 review 同时进入 Copilot AI Credits 和 GitHub Actions minutes，覆盖 Pro、Pro+、Business、Enterprise。 管理员需要核对 Actions entitlement / budgets / runner 设置，否则自动 PR review 会变成新的 CI 成本项。",
      "why_it_matters": "**双计费**：每次 review 同时进入 Copilot AI Credits 和 GitHub Actions minutes，覆盖 Pro、Pro+、Business、Enterprise。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "GitHub Changelog",
          "url": "https://github.blog/changelog/2026-04-27-github-copilot-code-review-will-start-consuming-github-actions-minutes-on-june-1-2026/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-microsoft-foundry-may-update",
      "title": "Microsoft Foundry 更新 agent 评测、成本归因和本地运行能力",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-30",
      "primary_entity": "Microsoft Foundry Blog",
      "event_type": "update",
      "object": "Microsoft Foundry 更新 agent 评测、成本归因和本地运行能力",
      "what_happened": "评测与追踪：Foundry 把托管/外部 agent traces、评测和项目视角成本归因放在同一组更新里。 本地能力：Foundry Local 1.1 带来实时 ASR、embedding、Qwen 3.5 Vision、WebGPU 插件；1.2 又补多语 ASR、ARM64、WinML 2.0。 成本归因只解释模型/项目使用，完整账单仍要结合 Azure Cost Management、Search、Storage、Key Vault 等资源看。",
      "why_it_matters": "**本地能力**：Foundry Local 1.1 带来实时 ASR、embedding、Qwen 3.5 Vision、WebGPU 插件；1.2 又补多语 ASR、ARM64、WinML 2.0。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "Microsoft Foundry Blog",
          "url": "https://devblogs.microsoft.com/foundry/whats-new-in-microsoft-foundry-may-2026/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-minimax-m3-model-page",
      "title": "MiniMax M3 发布，主打 coding、长上下文和原生多模态",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "MiniMax model page",
      "event_type": "launch",
      "object": "MiniMax M3 发布，主打 coding、长上下文和原生多模态",
      "what_happened": "三能力合一：M3 主打 open-weight、coding/agentic、1M context 和原生多模态，API 最低保证 512K context。 长任务数据：官方案例给出 ICLR 论文复现 12 小时、18 commits、23 figures，以及 FP8 GEMM 147 次提交、9.4x speedup。 开发者侧有 Token Plan、API、MiniMax Code 和待开源本地部署路径，价格/配额会直接影响 coding agent 使用成本。",
      "why_it_matters": "**长任务数据**：官方案例给出 ICLR 论文复现 12 小时、18 commits、23 figures，以及 FP8 GEMM 147 次提交、9.4x speedup。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "MiniMax model page",
          "url": "https://www.minimax.io/models/text/m3",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-nvidia-cosmos-3-physical-ai",
      "title": "NVIDIA Cosmos 3 面向 physical AI 推理、世界模型和动作模型",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "NVIDIA Developer Blog",
      "event_type": "update",
      "object": "NVIDIA Cosmos 3 面向 physical AI 推理、世界模型和动作模型",
      "what_happened": "架构变化：Cosmos 3 用 MoT 把 reasoner tower 和 generator tower 合并，输入可含文本、图像、视频、音频、动作。 模型规格：Nano 为 8B，面向工作站实时推理；Super 为 32B，面向 Hopper/Blackwell 数据中心部署。 NVIDIA 同步开放 6 类 synthetic datasets，并提供 BF16、FP8、NVFP4 NIM 路径，物理 AI 不只是视频生成。",
      "why_it_matters": "**模型规格**：Nano 为 8B，面向工作站实时推理；Super 为 32B，面向 Hopper/Blackwell 数据中心部署。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "NVIDIA Developer Blog",
          "url": "https://developer.nvidia.com/blog/develop-physical-ai-reasoning-world-and-action-models-with-nvidia-cosmos-3/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-nvidia-doca-ai-security",
      "title": "NVIDIA 用 DOCA In-Silicon Security 强化 agentic AI 基础设施",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "NVIDIA Developer Blog",
      "event_type": "update",
      "object": "NVIDIA 用 DOCA In-Silicon Security 强化 agentic AI 基础设施",
      "what_happened": "安全边界下沉：BlueField DPU 把监控、策略执行、遥测放到独立信任域，主机被攻破时仍可执行控制。 性能指标：DOCA 文章给出 runtime threat detection 最高 1,000x、网络/文件访问策略执行最高 800 Gb/s。 DOCA Argus、Vault、Flow 分别对应运行时威胁检测、文件级 zero-trust 访问和硬件加速网络策略。",
      "why_it_matters": "**性能指标**：DOCA 文章给出 runtime threat detection 最高 1,000x、网络/文件访问策略执行最高 800 Gb/s。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "NVIDIA Developer Blog",
          "url": "https://developer.nvidia.com/blog/advancing-ai-infrastructure-for-agentic-ai-with-nvidia-doca-in-silicon-security/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-nvidia-alpamayo-posttrain",
      "title": "NVIDIA Alpamayo 文章聚焦自动驾驶模型闭环后训练",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "NVIDIA Developer Blog",
      "event_type": "launch",
      "object": "NVIDIA Alpamayo 文章聚焦自动驾驶模型闭环后训练",
      "what_happened": "训练范式：AlpaGym 把 AlpaSim closed-loop rollouts 接入策略训练，让模型从自己动作造成的后果中学习。 工程依赖：教程要求 CUDA/cuDNN、NCCL、Redis、Git LFS、Hugging Face auth，并用 Hydra 配置 policy、scene、reward。 输出关注 mean reward、failure rates、policy loss、rollout throughput 和 checkpoint，可用于 AV 模型闭环验收。",
      "why_it_matters": "**工程依赖**：教程要求 CUDA/cuDNN、NCCL、Redis、Git LFS、Hugging Face auth，并用 Hydra 配置 policy、scene、reward。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "NVIDIA Developer Blog",
          "url": "https://developer.nvidia.com/blog/how-to-post-train-autonomous-vehicle-models-in-closed-loop-with-nvidia-alpamayo/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-alibaba-ticket-embedding",
      "title": "阿里云 Quick BI 推 ticket-based enhanced embedding 处理数据分享权限",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "Alibaba Cloud Blog",
      "event_type": "policy",
      "object": "阿里云 Quick BI 推 ticket-based enhanced embedding 处理数据分享权限",
      "what_happened": "权限模型：Quick BI ticket-based enhanced embedding 支持时间锁、访问次数锁和按角色个性化视图。 嵌入粒度：可嵌完整 dashboard、单个可视化组件、ad hoc query 和自助分析模块到 ERP、OA、CRM、App 或小程序。 治理看板跟踪授权用户、嵌入报表数、访问量和启用报表数，BI 分享从链接分发变成可运营服务。",
      "why_it_matters": "**嵌入粒度**：可嵌完整 dashboard、单个可视化组件、ad hoc query 和自助分析模块到 ERP、OA、CRM、App 或小程序。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "Alibaba Cloud Blog",
          "url": "https://www.alibabacloud.com/blog/sharing-data-without-risking-leaks-let-ticket-based-enhanced-embedding-strike-the-perfect-balance_603197",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-alibaba-lingyang-qwen-quickbi",
      "title": "Lingyang 在 Qwen Conference 展示 Quick BI 企业级数据方案",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-06-01",
      "primary_entity": "Alibaba Cloud Blog",
      "event_type": "update",
      "object": "Lingyang 在 Qwen Conference 展示 Quick BI 企业级数据方案",
      "what_happened": "会议场景：Qwen Conference Singapore 上，Quick BI 把 enterprise Agentic Analytics 讲成“Goals -> Inferences -> Actions”。 业务覆盖：文章列出电商运营、销售运营、制造运营、供应链、金融运营等场景，强调统一指标和权限边界。 案例称手工日报工作量下降约 90%、问题发现解决提速 10x，报表迁移人工校验工作量下降 50% 以上。",
      "why_it_matters": "**业务覆盖**：文章列出电商运营、销售运营、制造运营、供应链、金融运营等场景，强调统一指标和权限边界。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "Alibaba Cloud Blog",
          "url": "https://www.alibabacloud.com/blog/lingyang-debuts-at-the-qwen-conference-in-singapore-quick-bi-deconstructs-enterprise-grade-data-solutions-for-the-ai-era_603196",
          "type": "source"
        }
      ]
    }
  ],
  "huggingface_trending": [],
  "chinese_media_dynamics": [],
  "daily_tracking": [],
  "official_org_updates": [],
  "quality_status": {
    "status": "ok",
    "public_note": "Core discovery checks completed without blocking degradation."
  }
}