LoopX overall roadmap: long-running multi-Agent collaboration / LoopX 整体路线总纲 #4575
huangruiteng
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
LoopX 的整体方向 / Where LoopX is going
我们希望一个人能通过本地前端或 Lark 提出、修订和验收复杂目标,让持久管家协调多个长程 LoopX Agent,在本地 managed 与云端 runtime 上持续完成工作。每个 Agent 保有明确的目标、承诺与执行边界;管家组织、委托和综合,peer 之间也能直接求助、交接产物和独立复核。
LoopX should let a person express, revise and accept complex goals through the local frontend or Lark, while a persistent steward coordinates long-running LoopX Agents across local managed and cloud runtimes. Agents retain explicit commitments and execution boundaries. The steward organizes and synthesizes; peers can directly request help, hand off artifacts and review work.
Read the overall roadmap: 中文 · English
Execution tracker: #4574 — LoopX overall delivery program. The roadmap was merged in #4570.
一份总纲,覆盖完整产品 / One roadmap for the whole product
总纲连接原有 30 个主 RFC、13 条工作流、6 个产品组合里程碑和 7 张核心执行卡,并纳入 RFC 外的重要领域:
The roadmap connects all 30 pre-existing primary RFCs, 13 streams, 6 portfolio milestones and 7 core execution cards. It covers product/steward work, planning and peer collaboration; TS and shared authority; managed/attached/cloud hosts; frontend/Lark; materials, memory, budgets and scheduling; extensions, trust and operations; research, releases, developer/community experience and sustainable adoption.
协作能力如何证明 / What counts as collaboration
多 Agent 能力需要证明具体的工作关系:
Collaboration must prove parallel joins, versioned pipeline handoff, peer help/review, fenced execution-responsibility continuation and cross-host recovery. Message receipt, work adoption, effect commit, result acceptance and answer delivery are different facts. A directory, team card or a hundred registrations does not establish a working long-running team.
推进顺序 / Delivery order
可信基线 → 2–3 Agent 真实小团队 → 可恢复协作工作台 → 本地/云端共同工作 → 10/30/100+ 活跃 cohort 资格。
Trustworthy baseline → real 2–3-Agent team → recoverable workspace → local/cloud collaboration → qualified 10/30/100+ active cohorts.
最早的小团队里程碑就要求真实 peer handoff、依赖产物、至少两轮推进、故障恢复、用户补充与独立验收。百 Agent 是后续系统资格,不是先提高一个数量上限。成本、吞吐、恢复、人工介入和成果质量需要固定口径与完整失败分母。
The first team milestone already requires real peer handoff, dependent artifacts, two work cycles, recovery, owner correction and independent acceptance. Scale follows evidence rather than a larger cap. Measure cost, throughput, recovery, attention and accepted outcomes with fixed definitions and complete failure denominators.
采用路线也可以独立推进:已有 agent 系统先接入 L1 被动诊断 → L2 建议 → 经授权的 L3 seam → 可选 L4 控制面。它不必等待百 Agent;每次增加控制权都需要独立证据与授权。
Existing systems can adopt independently through L1 passive diagnostics → L2 advice → authorized L3 seams → optional L4 control-plane adoption. This need not wait for hundred-Agent scale; each authority increase needs its own evidence and authorization.
参与讨论与贡献 / How to participate
欢迎围绕以下问题给出可公开或合成的案例:
Share public-safe or synthetic examples of continuity failures, dependency stalls, valuable peer relationships, acceptance criteria, adoption friction, reproducible faults or adapter/user-journey contributions. Please exclude private transcripts, credentials and organizational context.
请先查看 执行总跟踪 #4574 及其已有领域 Issue;具体实现用有界 Issue/PR 跟踪,方向与取舍在这里讨论。现有技术方向讨论、管家与 handoff 讨论 和前沿研究讨论 继续作为相关领域入口。
Use bounded issues/PRs for implementation and this discussion for portfolio tradeoffs. The existing technical-directions, manager/handoff and frontier-research discussions remain useful domain entrypoints.
总纲合并意味着方向与执行路径可发现,不意味着所有 RFC 已接受、功能已交付、provider 已晋升或商业模式已验证。仓库总纲是当前计划来源;领域合同保留权威,实际 Todo/Issue/PR 保存执行证据。
Merging the roadmap makes direction and execution paths discoverable. It does not mean every RFC is accepted, every capability shipped, providers promoted or product-market fit established. Repository plans, domain authority and actual execution evidence remain separate.
All reactions