搜索“GitHub”

netes基盤で利用できる。既存のエージェントを動かせるほか、エージェントや必要なツールを組み込んだ独自のイメージも使える。 インストールしてOpenCodeを動かす例 公式の初回起動ガイドでは、コーディングエージェントのOpenCodeと、モデルAPIサービスのOpenRouterを使う例を紹介している。ほかにも、Piを動かす例やClaude CodeからGitHubへの書き込みを許可する例が公開されている。 ここでは、OpenCodeの例に沿って、導入から応答の確認までを示す。OpenCodeはローカルのサンドボックス内で動かし、AIモデルはOpenRouter経由で利用する。 この例では実行基盤にDockerを使う。LinuxにDocker Engine 28.0以降を導入・起動した環境と、OpenRouterのアカウント・APIキーを用意する。以下のコマンドは、サンドボックスを動かすLinuxホストのBashで順に実行する。サンドボックスのコンテナはOpenShellが作成・起動する。 まず、OpenShellのインストールスクリプトを実行すると、CLIやポリシーの検証ツールが

匹配字段正文

a session moves to another account, Claude rereads it and regenerates thinking. Roadmap: @mikeyk said Haiku 5.5 will “round out the family in the coming weeks.” Availability: It shipped day-one on the Claude Platform and Claude Code, along with a usage reset valid until Oct 22 ( @ClaudeDevs ). Third-party availability: GitHub Copilot in VS Code ( @code ) Arena Agent/Battle modes for WebDev, Text, Vision and Document ( @arena ) T3 Code, after @theo admitted it hadn’t been added to the catalog yet

匹配字段正文

Sol Ultra 发布之际，人们正争相在 之前寻找 Fable 5 的极限，请特别关注 Claude Mods（尤其是 cheatsheet）： Boris Cherny @bcherny Claude Mods 正在陆续上线。已经有人构建了一个 Claude 内运行的俄罗斯方块 mod 🤯 查看 issue 了解最新的社区更新、技术细节和更多酷炫演示 github.com/anthropics/cla… 下午5:30 · 2026年9月14日 · 60.3万次查看 312条回复 · 173次转发 · 2.72万次点赞 总的来说，这也是 Thariq 另一条病毒式传播推文的反面： Thariq @trq212 奇怪的是，有一个“赚大钱”的按钮，却没人按 （把你的 SaaS 变成无头模式，让智能体使用它，按交互收费，尤其是对企业） 下午10:39 · 2026年8月18日 · 55.9万次查看 333条回复 · 229次转发 · 6570次点赞 云端大脑，本地之手 并且试试 Claude Projects： Boris Cherny @bcherny Projects 不仅改变了我与

匹配字段正文

h benchmark, 100 English tasks, live Exa retrieval. Latency medians over 336–493 calls per model, each on its own provider's public API. Harness and per-call timing are public at: github.com/apoorvumang/retrieval-vs-recall Mercury 2 is the fastest model at every step of the pipeline, almost 2x faster than Gemini 3.1 Flash Lite on query planning, 4.7x faster than Claude Haiku 4.5, 10x faster than GPT-5 Mini. That per-step gap is where the end-to-end numbers in the next section come from: a pipeli

匹配字段正文

of these. Mercury 2 optimizes for all three. We evaluated Mercury 2 on PinchBench, the open-source benchmark built on top of OpenClaw — the fastest-growing open-source project in GitHub history, with 250K+ stars in under 60 days. Here's what we found. The result Mercury 2 sits in the upper-left of the chart: high task success rate and the fastest execution time in its performance class. 78% success rate — matching or exceeding GPT-5 Mini (75%), Gemini 2.5 Flash (71%), DeepSeek Chat (72%), and G

匹配字段正文

of your coding workflow Mar 30, 2026 Product We evaluated Mercury 2 on PinchBench, the open-source benchmark built on top of OpenClaw — the fastest-growing open-source project in GitHub history, with 250K+ stars in under 60 days. Mar 24, 2026 Product We evaluated Mercury 2 on PinchBench, the open-source benchmark built on top of OpenClaw — the fastest-growing open-source project in GitHub history, with 250K+ stars in under 60 days. Today, we're introducing Mercury 2 — the world's fastest reason

匹配字段正文

Infrastructure Z.ai uses slime, one of the best open-source RL training frameworks, as the post-training infrastructure throughout all GLM-5 models. We refer readers to the slime GitHub repo to learn more about its design and features, and here we highlight two features. Performance characteristics of RL rollouts are highly task-dependent, and each task comes with different tools and reward functions. To ensure system efficiency, Z.ai designed a server-based multi-task rollout orchestrator. Eac

匹配字段正文

through Claude Platform on AWS in North America. Give Claude Sonnet 5.5 a try in the Amazon Bedrock console, in Claude Platform on AWS, or explore the Getting Started notebooks on GitHub. About the authors Dani Mitchell Dani is a Senior Specialist Solutions Architect for Generative AI at AWS, working on go-to-market for Anthropic on Amazon Bedrock. He helps enterprises across the world design and deploy generative AI solutions using Anthropic’s models and capabilities on Amazon Bedrock to build

匹配字段正文

个包含的类型提供可用配额。如果 SageMaker 选择了不同的实例，您的每小时成本也可能发生变化。使用 --instance-types 将池限制为满足您配额、价格和性能要求的类型。 开始之前，您需要： 本演练使用美国东部（弗吉尼亚北部）区域。示例会根据 AWS REGION 构建 DLC 镜像 URI 和运行时端点。 git clone https://github.com/aws-samples/sagemaker-genai-hosting-examples.git cd sagemaker-genai-hosting-examples/03-features/bidirectional-streaming-vLLM-Omni 安装所需的 Python 包。创建虚拟环境并安装示例定义的版本。 python3.12 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt 配置示例。设置您的 SageMaker AI 执行角色。示例默认使用 us-east-1。传入 --re

匹配字段正文

geMaker 容量感知实例池可以按优先级顺序列出兼容的实例类型。只有在验证每个候选类型满足图像模型的 GPU 内存和性能要求后，才添加实例池。本示例将异步视频端点保持在固定实例类型上。 开始之前，你需要： 本演练使用美国东部（弗吉尼亚北部）AWS 区域。在部署 GPU 端点之前，请查看 SageMaker AI 定价。 git clone https://github.com/aws-samples/sagemaker-genai-hosting-examples.git cd sagemaker-genai-hosting-examples/03-features/vllm-omni-image-video 安装 Python 依赖项。创建虚拟环境并安装示例要求。 python -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt 部署图像和视频端点。设置您的 SageMaker AI 执行角色并运行部署脚本。该脚本固定 omni-sagemaker-cuda-v1.6，创

匹配字段正文

resilient to routine UI changes, while AgentCore Runtime provides the infrastructure to run workflows at scale. For the complete working implementation, see the sample repository: github.com/aws-samples/sample-nova-act-synthetic-monitoring Sarath is a Senior Gen AI/ML Specialist Solutions Architect at Amazon Web Services, where he helps enterprise customers design and deploy generative AI and machine learning solutions that deliver measurable business outcomes. He brings deep expertise in Agenti

匹配字段正文

自执行破坏性操作。4月，搭载Claude的Cursor编程智能体被曝删除软件企业PocketOS的生产数据库及相关备份，造成客户业务混乱，数据随后得到恢复。 数据与访问凭证外泄也是一种风险。OpenAI于9月披露，多个智能体在协作制作工作簿时，因无法互相读取本地文件，擅自将文件上传至公共托管平台。另一起案例中，内部模型为获取其他团队的数学证明材料，将研究员的GitHub访问令牌写入公开仓库，并拆分令牌以躲避密钥扫描。 还有一些行为涉及造假、欺骗与规避监督。OpenAI披露，部分模型会在用于延续任务的摘要中留下指令，要求编造缺失数据或隐瞒错误。 提示注入及恶意指令传播，则是另一条风险路径。6月公开的GitInject研究验证，恶意代码提交或配置文件可能诱导智能体泄露凭证、操纵审查判断。OpenAI于9月披露的信息也显示，恶意指令可以诱使智能体将其复制到后续邮件、文件或代码注释中。 独立网络安全研究机构DARKNAVY的AI安全研究员李忠睿对澎湃新闻记者表示，这一轮智能体安全事件与过去的大模型安全问题最大的区别，是风险正在从“看错东西、说错话”转变为“做错事”。 “过去的大模型安全，主要是

匹配字段正文

ect the rules. Agents write and run their own code, and when one path is blocked they look for another. In one of NVIDIA's tests, an agent that wasn't allowed to push code through GitHub's API simply switched to git instead. We've now seen countless examples of this behavior. For now, safety can't live only inside the agent. It has to be built around it. That's the concept behind OpenShell (open source, Apache 2.0): - the agent runs in a Linux sandbox (Landlock + seccomp): no root, no direct net

匹配字段正文

Anthropic integrated Claude Managed Agents with OpenShell and BlueField. SpaceXAI uses it for Cursor coding agents and Grok models. Salesforce connected OpenShell to Slack for approving agent permission requests. SAP is embedding OpenShell in the Joule Studio runtime. Red Hat, SUSE and Canonical are integrating it into their operating systems. The effort feeds the Open Secure AI Alliance , governed by the Linux Foundation. OpenShell and its skills are available on GitHub and the OpenShell docs .

匹配字段正文

r Docker and Kubernetes, and trusted middleware for identity services. Next Steps Start the quickstart to run an agent with OpenShell and configure its service access. Explore the GitHub repository to examine the code and contribute. Join the CNCF Slack openshell-dev channel to ask questions and connect with other developers. Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more AI agents can be given a goal, write code,

匹配字段正文

Stop building environment registries Every RL paper or framework uses its own way to find environments. Custom hubs, runtime registries, independent task datasets, or a GitHub list of tasks with a custom loader. This means that many of the published environments are siloed: if you publish an environment for one framework, users of the other three can’t load it. If you want to train on an environment from another framework or a new paper, you’ll need to port it by hand. We think this is the wrong

匹配字段正文

has its own leaderboard page: CWE-Bench-AA, DeepsecBench-AA, and CyberGym-E2E-AA; Stirrup, the open-source agent harness used for every Artificial Analysis Cyber Index task, is on GitHub Our development roadmap Security teams are already choosing models to support their defensive work, so we are launching the Artificial Analysis Cyber Index and the Cyber Index Alliance now, with three evaluations, rather than waiting to cover every capability. We will review the Index as models improve and add e

匹配字段正文

e state of the art for November. Claude still couldn’t really draw a bicycle! The GPT-5.1 bicycle frame is pretty crap too. Also in November, we had the first commit to an obscure GitHub repository called “Warelay”. We’ll come back to this repository shortly. And then there were the December holidays, and individual developers took some time off and many started tinkering with these new coding agent model combinations... and it began to dawn on us quite how much they could do that they couldn’t

匹配字段正文

needs no access to model internals. A Global Selection step picks the best video across all iterations, not simply the last one. Google built 3 new benchmarks. GenAD-Bench has 400 ad scenarios across 200 fictional products from 50 brands. HardContinuityBench stresses scene reappearances and prop state changes. LVBench-C has 120 scenarios where key assets vanish for at least 10 segments before returning. Sources: linked papers, project pages, and GitHub repositories. Verified September 27, 2026.

匹配字段正文

lse positives under our monitor’s scope of real-world harm. These involve: Attempting to tamper with a task’s scorer (9/10 and 6/10 suspicion); Attempting (and failing) to push to GitHub from a sandbox with no credentials (9/10 suspicion); Printing synthetic credentials given by the task into the transcript (10/10 and 9/10 suspicion) We also looked at the most concerning cases rated below our 3/10 threshold (i.e. presumed true negatives) by filtering to capable agents (opus-4.7 / gpt-5.4 and abo

匹配字段正文