按主题看 AI
按公司与模型、技术方向、内容形态浏览 15 个主题，持续汇集近期焦点与精选。
公司与模型
按厂商与模型系追踪：谁发了什么、又赢了哪一局
- DotsIndependent tracking of Dots documentation, updates, workflows, templates and publicly verifiable signals.查看 0 条精选 →
- AstraAstra releases, models, API notes, coding and computer-use capabilities, with evidence labels.查看 0 条精选 →
- AEON WatchA source-led timeline for AEON-related public evidence: confirmed facts, code signals, community findings and analysis.查看 0 条精选 →
- OpenAIOpenAI first-party releases and independent reporting relevant to agents, APIs, models and long-running workflows.查看 0 条精选 →
技术方向
按技术领域深挖：Agent、多模态、具身智能……
- Always-On AgentsSystems that continue work across time, triggers and context instead of ending at one chat turn.查看 0 条精选 →
- Persistent AgentsLong-horizon agents with state, schedules, tools, approvals and explicit stop conditions.查看 0 条精选 →
- Agent RuntimeThe execution loop, tool boundary, memory, observability and approval layer around an agent.查看 0 条精选 →
- Agent MemoryState and retrieval patterns that let an agent carry useful context across runs without hiding provenance.查看 0 条精选 →
- Computer UseAgents operating browsers, desktops and software surfaces with explicit permissions and recovery paths.查看 0 条精选 →
- MCP & toolsTool and context protocols that connect agents to systems while keeping capabilities inspectable.查看 0 条精选 →
内容形态
按内容类型浏览：论文、教程、观点、政策……
- Agent TemplatesReusable Agent Work Specifications with triggers, actions, approvals, outputs and stop conditions.查看 0 条精选 →
- GuidesEvergreen explainers for designing, evaluating and operating persistent agents.查看 0 条精选 →
- Open Source AgentsOpen repositories and runtime building blocks that can be inspected, forked or self-hosted.查看 0 条精选 →
- Agent SecurityPermissions, human approval, prompt injection boundaries, auditability and operational failure modes.查看 0 条精选 →
- Human ApprovalApproval gates and stop rules that keep long-running work accountable to people.查看 0 条精选 →