01
运行底座Operating Base
身份、能力、协作和结算要在真实任务中可识别、可调用、可记录。Identity, capability, collaboration, and settlement need to be identifiable, callable, and recordable in real tasks.
BLP / 03AIOS 蓝图AIOS Blueprint
AIOS 的下一步由什么决定？What decides the next step for AIOS?
AIOS 的蓝图，是让机制在真实协作中留下证据，再由证据推进下一阶段。The AIOS blueprint lets operating mechanics leave evidence in real collaboration, then uses that evidence to move the system forward.
身份、能力、协作和结算先在真实任务中成立；能力被持续创建、调用、改进和复用之后，才进入更完整的能力网络与能力经济判断。Identity, capability, collaboration, and settlement must first hold in real tasks. Only when capabilities are repeatedly created, invoked, improved, and reused can AIOS judge the move toward capability networks and capability economy.
从机制到阶段判断From mechanics to stage judgment
04 项04 SIGNALS02
可信协作Trusted Collaboration
当前先验证不同主体的能力能被持续调用、产生结果并形成可复用反馈。The current stage first validates whether capabilities across subjects can be invoked repeatedly, produce outcomes, and form reusable feedback.
03
切换证据Transition Evidence
稳定供给、可靠调用、持续改进、跨场景需求和参与者持续性共同决定阶段切换。Stable supply, reliable invocation, continued improvement, cross-scenario demand, and participant persistence decide stage transition together.
04
长期约束Long-term Constraints
基础设施优先、从协作切入、开放网络优先；真实证据出现前不固化未来机制。Infrastructure first, collaboration first, open network first; future mechanisms do not harden before real evidence exists.
01运行闭环Operating Loop
一条可复用关系，必须先在真实任务里被写下。A reusable relationship has to be written in real tasks first.
能力带着身份和调用条件进入，需求在明确上下文中找到可信能力，执行穿过智能体、设备、服务与空间，结果和反馈回到下一次协作。A capability enters with identity and invocation conditions; demand finds trusted capability in context; execution crosses agents, devices, services, and spaces; outcomes and feedback return to the next collaboration.
01
能力声明Capability Declaration
能力必须带着身份、边界、权限和调用条件进入 AIOS，才可能被发现和调度。A capability has to enter AIOS with identity, boundaries, permissions, and invocation conditions before it can be discovered and orchestrated.
02
需求发现Demand Discovery
需求在明确上下文中找到可信能力，而不是依赖临时拼接或人工记忆。Demand finds trusted capabilities inside explicit context instead of relying on temporary assembly or human memory.
03
现场调用Field Invocation
一次调用要穿过智能体、设备、服务和空间，在真实约束中完成任务。One invocation passes through agents, devices, services, and spaces, then completes the task under real constraints.
04
结果回流Outcome Return
结果、贡献和反馈被记录后，才会留下边，并推动能力再次改进。Only recorded outcomes, contributions, and feedback leave an edge and improve capabilities for the next use.
02当前阶段Current Stage
当前要证明的不是完整网络，而是可信协作可以复用。The current proof is not a complete network, but reusable trusted collaboration.
AIOS 当前处于验证可信协作的阶段：不同主体的能力能否在真实场景中持续调用，结果与贡献能否被记录，反馈能否推动再次使用。AIOS is currently validating trusted collaboration: whether capabilities across subjects can be repeatedly invoked in real scenarios, whether outcomes and contributions can be recorded, and whether feedback can drive reuse.
01
能力被创建Capability Created
开发者、智能体、设备或空间能力能够被创建、接入，并留下可识别边界。Developer, agent, device, or space capabilities can be created, connected, and given identifiable boundaries.
02
真实场景调用Real Invocation
能力不是停在目录中，而是在真实任务和现场约束中被调用。Capabilities do not stop at a catalog; they are invoked in real tasks and field constraints.
03
结果可记录Outcome Recorded
结果、贡献、失败位置和恢复过程需要被记录，才可能成为下一次协作的依据。Outcomes, contributions, failure location, and recovery need to be recorded before they can inform the next collaboration.
04
反馈再使用Feedback Reused
反馈必须推动能力改进，并让改进后的能力再次进入任务。Feedback must improve the capability and bring the improved capability back into tasks.
03切换证据Transition Evidence
进入能力网络，只看真实运行证据。The move into capability networks depends only on operating evidence.
阶段变化不由年份、叙事需要或未来方案完整度触发。只有供给、调用、改进、需求和参与持续成立，下一阶段才有讨论基础。Stage transition is not triggered by a date, narrative pressure, or a complete future proposal. The next stage becomes discussable only when supply, invocation, improvement, demand, and participation keep holding.
01
稳定供给Stable Supply
能力能够被持续创建、维护和复用；未满足时继续降低创建与维护成本。Capabilities can be created, maintained, and reused continuously; if not, creation and maintenance costs remain the focus.
02
可靠调用Reliable Invocation
跨主体调用在真实场景中稳定完成，失败可定位、可恢复。Cross-subject invocation completes reliably in real scenarios, with failures locatable and recoverable.
03
持续改进Continued Improvement
使用反馈能够推动能力版本继续改进，而不是只留下单次结果。Usage feedback improves capability versions instead of leaving only one-off outcomes.
04
跨场景需求Cross-scenario Demand
能力发现与复用开始超出单一场景，才具备网络级讨论价值。Capability discovery and reuse need to move beyond one scenario before network-level mechanisms are worth discussing.
05
参与者持续性Participant Persistence
开发者和场景方愿意持续参与，网络投入才不会变成一次性展示。Developers and scenario owners need to keep participating, or network investment becomes a one-off display.
04长期约束Long-term Constraints
长期选择把 AIOS 保持在基础设施轨道上。Long-term choices keep AIOS on the infrastructure track.
AIOS 不靠提前堆满未来功能取胜，而是持续守住三条边界：基础设施优先、从协作问题切入、开放网络优先。AIOS does not win by prebuilding every future function. It keeps three boundaries: infrastructure first, collaboration as the entry problem, and open networks first.
01
基础设施优先Infrastructure First
长期价值来自让智能体、能力、设备与场景可靠协作的基础设施，而不是单一终端应用流量。Long-term value comes from infrastructure that lets agents, capabilities, devices, and scenarios collaborate reliably, not from single-application traffic.
02
从协作切入Collaboration First
当前从可信协作切入；是否进入能力网络或能力经济，只由真实证据判定。The current entry point is trusted collaboration; any move toward capability networks or capability economy is decided only by real evidence.
03
开放网络优先Open Network First
能力应能跨主体被发现、调用和验证；阶段性中心入口不能固化为封闭终局。Capabilities should be discoverable, callable, and verifiable across subjects; temporary centralized entry points must not become a closed end state.
04
不提前固化No Premature Hardening
在真实证据形成前，不提前固化未来阶段的定价、分配、治理或网络级方案。Before real evidence exists, future pricing, distribution, governance, or network-level mechanisms do not harden.
从证据进入下一步From Evidence to Next Step
先看机制如何运行，再看一次任务、一个产品或一个长期问题如何承接。Start with how the mechanics operate, then follow a task, product, or long-term question.
查看系统机制，理解身份、能力、协作和结算如何成立；查看能力执行，理解一次任务如何被组织；回到产品体系或前沿研究，继续看支撑关系与长期问题。Use system mechanics to understand identity, capability, collaboration, and settlement; use runtime execution to see one task being organized; continue into products or research for support relationships and long-term questions.