Tag
An open-source project announcement for FlowX, an MCP-based workflow compiler and evolution engine that lets agents turn natural-language requirements into executable workflows, extract reusable skills from execution experience, and self-improve over time. The author is seeking collaborators in AI agents, workflow engines, MCP, and self-evolving systems.
Vercel introduces Vercel Board, an experimental secure message board described as "headless Slack" for humans and AI agents, designed to be model-, harness-, and platform-agnostic for building agent swarms and software factories.
An open-source repo presenting a complete course on harness engineering for coding agents — 14 lectures, 8 projects, exercises, and reusable templates covering instructions, persistent state, verification, session lifecycle, maker-checker loops, and graph-based workflows.
The post surveys the growing ecosystem of API-based tools built for AI agents — covering email, phone, browsers, cloud sandboxes, payments, memory, search, and social data — noting that workflows that once required custom scripts are now available as composable services, while questioning real-world reliability beyond demos.
DeerFlow is given a secure Docker-based "computer" sandbox that can execute commands, manage files, and run long tasks, bundling Browser, Shell, File, MCP and VSCode Server in a single container with uploads, workspace, and outputs folders.
Databricks Vice President Nikita Shamgunov believes Claude Code is just the beginning. The AI software factories of the future will require entirely new agent infrastructure, and state management (such as database state) will be one of the hardest problems to solve.
Alibaba Cloud has open-sourced the 'AI Agent Handbook', a practical guide to the full enterprise Agent lifecycle covering architecture, building, running, governance, and tuning. It includes 2026 Agent developer survey data, highlighting the key challenges and engineering practices for moving enterprise Agents from pilots to governed production.
Mintive launches with $11M raised (led by Cowboy Ventures) to build agent infrastructure giving every AI agent its own identity, permissions, logging traces, and token budgets as companies shift from chatbots to autonomous agents.
Aaron Levie argues that autonomous, domain-specific agents represent the emerging form factor for AI, driven by better tool/computer use, improving model capabilities, and mature agent harnesses and infrastructure — positioning them as a major leap beyond simple LLM wrappers for enterprise workflows.
Perplexity 开源了其代理沙箱平台 SPACE,该平台是 Computer 背后的安全基础设施,为开发者提供隔离执行、状态恢复和网络/凭证控制等生产级代理运行时基础。
OpenHands 宣布成为 NVIDIA Nemotron 3.5 Super VL 的发布合作伙伴,将这款开放多模态模型引入长期运行的软件工程智能体工作流,使智能体能够理解截图、仪表盘、文档和图表等非文本上下文。
BOSS Console 在 Product Hunt 发布,定位为面向常驻 AI 代理的轻量级操作系统,支持创建带真实工具的代理、通过 fluck.ai 聊天并访问实时会话。
BabyAGI 创始人 Yohei Nakajima 分享了他对组织内 agent 协作未来的看法,认为未来将是一个面向 agent 与人类协作的“CompanyOS”(公司操作系统),用于存储规则、流程和决策,而非一个集中式的“公司大脑”,并将其类比为持续更新的“flipped playbook”而非瀑布式的规则制定。
AgentKit by Scalekit is an MCP gateway designed for production agents that act on a user's behalf, handling the infrastructure layer for agent tool access.
公司宣布其 Agent 平台 2.0 大幅简化:砍掉 pods、workflow、memory bank 和自研 SDK,改为兼容 OpenAI Agents API,支持任意模型(Claude、Gemini、开源权重、OpenRouter 或用户自带 key),并聚焦于故障恢复、精确成本日志和严格按用户指定模型运行。
推荐 Yifan Xu 开源的 Agent Infra 项目 LayerFS,通过 CAS + CDC + COW 实现低开销的工作区分叉,让多个 Agent 能从同一状态并行独立探索,可直接丢弃失败分支并保留有价值的结果,预计10天内发布。
推文整理了 Agent Infra(智能体基础设施)这一新兴领域的 12 项核心技能,涵盖 Agent 运行时、上下文与状态管理、任务队列、多智能体编排、持久化执行、沙箱环境、MCP 工具协议、权限安全、可观测性与评估等方面,为有意进入该方向的开发者提供学习路线图。
LangChain introduced Managed Deep Agents at the Interrupt NYC conference, bundling the agent harness with infrastructure to simplify authentication, memory, and channel integrations, while adding a pre-built web search tool, Harbor evaluation integration, and Parallel tool integration — allowing teams to focus on business logic.
Monid 被称作「Agent 世界的 OpenRouter」,聚合了 2,443 个工具与 API,覆盖社交媒体数据、搜索、浏览器自动化、AI 内容生成、企业情报等几乎所有常见功能,推文列举了其中 30 个代表性集成。
EmpirioLabs released Aplomb 1, an open-weights 5.3B decision model built on Qwen3.5-4B that handles 1M-token text, image, video and audio inputs, returns probabilities for tool arguments, and tops the Decision Index board among sub-4B models while claiming leading scores on 8 of 38 benchmarks up to 5.3B.