diff --git a/README.md b/README.md
index 365edecac2..ba53822001 100644
--- a/README.md
+++ b/README.md
@@ -9,13 +9,13 @@
Tell the Agent you already use what you want done. LoopX keeps the goal, work,
needed decisions and verifiable results durable behind the scenes, so the next
session can continue without making you learn LoopX internals first.
-Work with Codex, Claude Code, DeepSeek Harness and other runtimes. Advanced docs still expose the local-first architecture, ownership model and research workbench.
+An open, harness-neutral control plane for 24/7 always-on agents across Codex, Claude Code, DeepSeek Harness and other runtimes—scaling from local SQLite workflows to multi-agent teams.
[](https://github.com/loopx-project/loopx/releases/latest) [](LICENSE) [](docs/architecture/rfcs/typescript-control-plane-migration-v0.md) [](pyproject.toml)
[](docs/public-private-boundary.md) [](docs/product/release-readiness.md) [](https://github.com/loopx-project/loopx/graphs/contributors) [](https://discord.gg/XmGgQyCFZd) [](https://dsh-plugin.org/plugins/loopx-project/loopx)
-[Get started](#quick-start) · [Why LoopX](#why-loopx) · [See it in action](#see-loopx-in-action) · [Evidence](#evidence) · [Contribute](#contributing) · [Docs](https://loopx-project.github.io/loopx/docs/) · [简体中文](README.zh-CN.md)
+[Get started](#quick-start) · [Why LoopX](#why-loopx) · [24/7 Always-On](#always-on-agents) · [See it in action](#see-loopx-in-action) · [Evidence](#evidence) · [Contribute](#contributing) · [Docs](https://loopx-project.github.io/loopx/docs/) · [简体中文](README.zh-CN.md)
@@ -54,6 +54,7 @@ run? [Choose a task below](#try-loopx).
not news. Keep both languages aligned, newest first, using verified publication
dates (UTC). -->
+- **2026-11-01 · 📘 [Progressive context delivery for 24/7 agents](https://loopx-project.github.io/loopx/blog/context-delivery/)** — Compaction-cycle structured deltas and progressive views cut control-plane bytes across 213 model-facing wakes (`32.5%` single-layer progressive view; `3.0%` body / `5.0%` CLI stacked first-read entry across 112 in-cycle continuations) while preserving lossless baseline recovery ([47-day freshness postmortem](https://loopx-project.github.io/loopx/blog/five-freshness-incidents/)).
- **2026-10-25 · 🤝 [Long-horizon agent research: collaborators welcome](https://loopx-project.github.io/loopx/blog/edgebench-feedback-and-memory/)** — Exploring **breakthroughs in long-horizon outcomes under equal budgets**, through feedback, replanning and reproducible evaluation on EdgeBench and real-world tasks. Academic labs and industry teams are welcome to collaborate, comparing outcomes and total cost. [Discuss collaboration](https://github.com/loopx-project/loopx/discussions).
- **2026-10-20 · 🚀 [LoopX 2.0](https://github.com/loopx-project/loopx/releases/tag/v2.0.0)** — A shared TypeScript state layer and long-horizon research workbench. [Migration guide](docs/product/releases/v2.0.0.md) · [Engineering story](https://loopx-project.github.io/loopx/blog/typescript-control-plane-refactor/).
- **2026-10-14 · 🤖 [Agentic software](https://loopx-project.github.io/loopx/blog/semantic-software-development/)** — Connecting goals, evidence and human feedback to engineering delivery.
@@ -96,6 +97,7 @@ protected actions.
| --- | --- |
| Keep a coding or research agent working across sessions | [Install and start](#quick-start) |
| Manage personal projects, schedules and decisions in one place | [Personal Agent Workspace](#meet-the-personal-agent-workspace) |
+| Wire up a 24/7 always-on agent without hand-rolling infra | [24/7 Always-On Agent Control Plane](#always-on-agents) |
| Let agents collaborate and deliver verifiable results | [Agent collaboration guide](docs/product/use-cases/cross-runtime/README.md) |
LoopX keeps the durable work state in one compact layer:
@@ -142,6 +144,37 @@ publishing, production writes, and final ownership stay with the human.
+
+
+### Building 24/7 Always-On Agents Without Hand-Rolling the 5 Subsystems
+
+[Anthropic's *Building always-on agents*](https://claude.com/resources/articles/building-always-on-agents)
+notes that effective persistent agents spend most of their wall-clock life
+asleep at zero tokens and wake for bounded runs across five surrounding
+subsystems: **Event Inbox**, **Timers**, **Memory**, **Permissions**, and an
+**Outbox**. In practice, teams often re-build fragile glue—cron scripts, JSON
+queues, prompt compaction, and ad-hoc approval checks—for every agent host, only
+to hit silent staleness, duplicate writes, or noisy status spam by week two.
+
+LoopX packages that open, harness-neutral control layer so the runtimes you
+already use (Codex CLI, Claude Code, DeepSeek Harness, or custom runners) can
+operate 24/7 with explicit durability and human gates:
+
+| Always-on primitive | LoopX control-plane subsystem | What it prevents in multi-week runs |
+| --- | --- | --- |
+| **Event Inbox** | Single-flight **Goal / Todo / Hard Lease** queue, async app inbox, and connected Lark / GitHub listeners | Duplicate workers mutating the same repo state; dropped human inputs mid-run |
+| **Timers & Wakeups** | **Guarded Heartbeat** and safe recurring presets (`daily-triage`, `changelog-draft`, `pr-watch`; inspect via `loopx preset list`) | Blind cron loops burning tokens when nothing changed or a gate is still waiting on a human |
+| **Memory & Compaction** | [`Decision Context`](loopx/capabilities/decision_context/README.md) source tracking, compaction-cycle **Progressive + Delta** [`context_delivery`](https://loopx-project.github.io/loopx/blog/context-delivery/) (`32.5%` view across 213 wakes; `3.0% / 5.0%` first-read entry across 112 in-cycle wakes), and optional [`Reward Memory`](loopx/capabilities/reward_memory/README.md) | Silent source staleness ([5 incidents across 253 turns / 47 days](https://loopx-project.github.io/loopx/blog/five-freshness-incidents/)); compounding prompt replay or broken deltas across compaction |
+| **Permissions & Deferred Calls** | Explicit **owner, safety, publication, and private-data gates**, isolated worktrees, and review-ready handoffs | Autonomous cron runs publishing externally, forcing merges, or leaking private context |
+| **Outbox** | Shared Workspace & Lark Goal/Decision/Evidence views surfacing genuine completions, blockers, and open gates | Fragmented terminal logs and high-noise pings that force humans to babysit runs |
+
+**Solo local to team & organizational scale—on one set of contracts:**
+
+| Scale | How LoopX runs | Authority, collaboration & governance posture |
+| --- | --- | --- |
+| **Solo developer & local agent** | `python3 -m pip install --upgrade loopx` with zero external infrastructure; connect your existing CLI/Desktop runtime | Canonical local **SQLite** store with hard leases (plus existing File setups), surviving restarts and compaction at zero idle tokens |
+| **Engineering team & enterprise path** | Peer workers claim todos and return reviewed artifacts on shared Goals via Desktop Workspace, GitHub, and Lark ([three-agent demo](examples/collaboration-delivery/README.md)) | Preserves worker ownership and human approval gates; extends to staged **PostgreSQL** remote authority ([provider](loopx/control_plane/coordination/postgresql_authority_store.ts)) and planned org governance—RBAC, SSO, quotas, signed audit—in the [adoption & SaaS roadmap](docs/product/roadmaps/saas-opportunity-assessment.md) |
+
## See LoopX in Action

@@ -550,11 +583,13 @@ owner. See [Architecture](docs/architecture.md) and
### Personal Agents, Teams, and Self-Improving Workflows
-[Meta Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/)
-and [Grok Bot](https://x.ai/news/introducing-grok-bot) make persistent personal
-agents and delegated work a familiar product idea. LoopX approaches that space
-as an open, provider-neutral control plane for agents you already run—not as a
-hosted replacement for either product.
+[Meta Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/),
+[Grok Bot](https://x.ai/news/introducing-grok-bot), and
+[Anthropic's *Building always-on agents*](https://claude.com/resources/articles/building-always-on-agents)
+make persistent personal agents and sleep/wake event-driven runs a familiar product
+and architecture model. LoopX approaches that space as an open, provider-neutral
+[24/7 control plane](#always-on-agents) for agents you already run—not as a hosted
+replacement for either product.
- **Personal agent:** use the shipped Workspace and connected Lark surfaces to
inspect goals, steer work and resolve decisions. The
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 6a28ff47c2..32c2ef80de 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -8,13 +8,13 @@
告诉你已经在用的 Agent 想完成什么。LoopX 在幕后持久保留目标、
工作、审批与证据,让下一轮会话可以继续,而不要求你先学会 LoopX 内部机制。
-连接 Codex、Claude Code、DeepSeek Harness 等运行时。高级文档仍会展开本地优先语义内核、ownership 模型与研究工作台。
+面向 7×24 常驻在线 Agent 的开源、Harness 中立控制面——连接 Codex、Claude Code、DeepSeek Harness 等运行时,从个人本地 SQLite 平滑延伸到多 Agent 团队协作。
[](https://github.com/loopx-project/loopx/releases/latest) [](LICENSE) [](docs/architecture/rfcs/typescript-control-plane-migration-v0.zh-CN.md) [](pyproject.toml)
[](docs/public-private-boundary.md) [](docs/product/release-readiness.md) [](https://github.com/loopx-project/loopx/graphs/contributors) [](https://discord.gg/XmGgQyCFZd) [](https://dsh-plugin.org/plugins/loopx-project/loopx)
-[快速上手](#快速开始) · [为什么需要 LoopX](#为什么需要-loopx) · [看看实际使用](#看看-loopx-如何工作) · [证据与案例](#证据) · [参与贡献](#贡献) · [文档](https://loopx-project.github.io/loopx/docs/) · [English](README.md)
+[快速上手](#快速开始) · [为什么需要 LoopX](#为什么需要-loopx) · [7×24 常驻架构](#always-on-agents) · [看看实际使用](#看看-loopx-如何工作) · [证据与案例](#证据) · [参与贡献](#贡献) · [文档](https://loopx-project.github.io/loopx/docs/) · [English](README.md)
@@ -50,6 +50,7 @@ LoopX 会按需把它变成可见的任务、决策门和证据。Claude Code
not news. Keep both languages aligned, newest first, using verified publication
dates (UTC). -->
+- **2026-11-01 · 📘 [面向 7×24 Agent 的渐进式上下文交付](https://loopx-project.github.io/loopx/blog/zh/context-delivery/)** — 受 Compaction 周期约束的结构化差分与按需视图在 213 次模型唤醒中将单层渐进载荷降至原文 `32.5%`,112 次同周期续跑的叠加首读入口降至 `3.0%` 正文 / `5.0%` 含封装,同时保留压缩失配时原子回切完整基线的能力(延伸阅读:[47 天保鲜事故复盘](https://loopx-project.github.io/loopx/blog/zh/five-freshness-incidents/))。
- **2026-10-25 · 🤝 [长程 Agent 研究:欢迎学术与企业合作者](https://loopx-project.github.io/loopx/blog/zh/edgebench-feedback-and-memory/)** — 探索**等预算下的长期效果突破**:围绕 EdgeBench 和真实长程任务,研究反馈、重规划与可复现评测,同时核对效果与完整成本。欢迎学术与企业团队合作。[交流合作](https://github.com/loopx-project/loopx/discussions)。
- **2026-10-20 · 🚀 [LoopX 2.0](https://github.com/loopx-project/loopx/releases/tag/v2.0.0)** — 共享 TypeScript 控制面与长程研究工作台。[迁移指南](docs/product/releases/v2.0.0.md#中文指南) · [重构历程](https://loopx-project.github.io/loopx/blog/zh/typescript-control-plane-refactor/)。
- **2026-10-14 · 🤖 [Agentic Software](https://loopx-project.github.io/loopx/blog/zh/semantic-software-development/)** — 连接目标、证据、人的反馈与工程交付。
@@ -82,6 +83,7 @@ LoopX 会按需把它变成可见的任务、决策门和证据。Claude Code
| --- | --- |
| 让编程或研究 Agent 跨会话持续干活 | [安装并开始](#快速开始) |
| 在一个入口管理个人项目、定时任务与待决事项 | [个人 Agent 工作区](#认识个人-agent-工作区) |
+| 不从零手搓基础设施,直接搭建 7×24 常驻在线 Agent | [7×24 常驻 Agent 控制面](#always-on-agents) |
| 让多个 Agent 分工协作,交付可验收的结果 | [Agent 协作指南](docs/product/use-cases/cross-runtime/README.md#中文指南) |
LoopX 把长期控制状态留在同一层紧凑状态里:
@@ -126,6 +128,29 @@ LoopX 不是生产自动化控制器。危险权限、生产写入、公开发
+
+
+### 搭建 7×24 常驻在线 Agent:无需从零手搓 5 大底层组件
+
+[Anthropic 在《Building always-on agents》中指出](https://claude.com/resources/articles/building-always-on-agents),真正能长期稳定运行的常驻 Agent 大部分时间其实在**零 Token 休眠**,只在事件到来时被唤醒执行一轮有界工作,围绕 **Event Inbox、Timers、Memory、Permissions、Outbox** 五个底层原语运转。现实工程中,团队常常为每个宿主机重复手搓 cron 脚本、JSON 队列、Prompt 压缩和临时审批检查,运行几周后还是会在状态过期、重复写入或消息刷屏中翻车。
+
+LoopX 把这套**开源、Harness 中立**的控制面直接做成可运行基础设施,让 Codex CLI、Claude Code、DeepSeek Harness 或自研运行时共享持久状态与人工决策门:
+
+| 常驻 Agent 原语 | LoopX 对应控制面子系统 | 解决跨周运行中的哪类故障 |
+| --- | --- | --- |
+| **Event Inbox(事件收件箱)** | Single-flight **Goal / Todo / Hard Lease** 队列、异步对话收件箱,以及飞书 / GitHub 事件监听 | 多 Worker 并发踩踏同一份仓库状态;运行中途用户消息丢失 |
+| **Timers(定时与唤醒)** | **Guarded Heartbeat** 与安全周期预设(`daily-triage`、`changelog-draft`、`pr-watch`,可用 `loopx preset list` 只读查看) | 盲目 cron 在零增量或人门尚未批准时反复唤醒模型白烧 Token |
+| **Memory(记忆与压缩恢复)** | [`Decision Context`](loopx/capabilities/decision_context/README.md) 来源追踪、受 Compaction 周期约束的 **Progressive + Delta** [`context_delivery`](https://loopx-project.github.io/loopx/blog/zh/context-delivery/)(213 次唤醒中单层渐进降至原文 `32.5%`,112 次同周期续跑首读降至原文 `3.0% / 5.0%`),以及可选 [`Reward Memory`](loopx/capabilities/reward_memory/README.md) | 数据源过期却一路亮绿灯([47 天 / 253 轮中的 5 次保鲜事故](https://loopx-project.github.io/loopx/blog/zh/five-freshness-incidents/));上下文压缩后全量重放或差分悬空 |
+| **Permissions(权限与延后批准)** | 显式 **Owner / 安全 / 公开发布 / 私有数据 Gate**、隔离 Worktree 与可评审交付交接 | 定时任务在无人确认时私自对外发布、强行合并或越界读写 |
+| **Outbox(低噪成果回传)** | 本地工作区与飞书共享同一份 Goal / 决策 / 证据视图,仅在真正完成、阻塞或待决时打扰人类 | 碎片化终端日志与各端高频无效推送,避免人肉盯盘 |
+
+**个人开发者本地运行 → 工程团队与组织级扩展,同一套协议无缝衔接:**
+
+| 使用形态 | LoopX 运行方式 | 权威存储、协作与治理形态 |
+| --- | --- | --- |
+| **个人开发者 & 本地 Agent** | `python3 -m pip install --upgrade loopx` 零外部依赖直接运行,接入已有 CLI / 桌面端 | 默认本地权威 **SQLite** + hard lease(兼顾既有 File 配置),跨重启、跨压缩保留现场且闲置零 Token |
+| **工程团队 & 组织级扩展** | 多 Agent 认领 Todo、基于同一 Goal 经工作区 / GitHub / 飞书协同交付与互审([三 Agent 协作示例](examples/collaboration-delivery/README.md)) | 保留各自责任边界与人工审批门;支持向预置的 **PostgreSQL** 远程共享权威([源码实现](loopx/control_plane/coordination/postgresql_authority_store.ts))以及[组织治理路线图](docs/product/roadmaps/saas-opportunity-assessment.zh-CN.md)中的 RBAC、SSO 与审计能力平滑演进 |
+
## 看看 LoopX 如何工作

@@ -490,10 +515,12 @@ provider 的打包和生命周期,不是另一个控制面 owner。详见
### 个人 Agent、数字团队与自我改进
-[Meta Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/)
-与 [Grok Bot](https://x.ai/news/introducing-grok-bot) 正在让“把工作交给常驻的个人 Agent”
-变得更容易理解。LoopX 选择的是开放、Provider-neutral 的控制面路线,
-连接你已经在用的 Agent,而不是提供这两款托管产品的等价替代。
+[Meta Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/)、
+[Grok Bot](https://x.ai/news/introducing-grok-bot) 以及
+[Anthropic《Building always-on agents》](https://claude.com/resources/articles/building-always-on-agents)
+正在让“把工作交给 7×24 常驻个人 Agent”与“休眠—事件唤醒”成为越来越清晰的产品与架构共识。LoopX
+选择的是开放、Provider-neutral 的[7×24 控制面](#always-on-agents)路线,连接你已经在用的 Agent,
+而不是提供这两款托管产品的等价替代。
- **个人 Agent:**通过已发布的工作区与关联飞书入口查看目标、调整方向、处理决策。
[常驻管家与语义交接 RFC](docs/architecture/rfcs/capable-manager-semantic-handoff-v0.zh-CN.md)