Portable work units可携带的工作单元
Every mission carries its outcome, scope, context, acceptance gates, and write boundaries. It can move without losing what matters.
每项任务都携带目标、范围、上下文、验收门槛与写入边界,流转时不会丢失关键条件。
Matrix platform · Work becomes operational
Matrix gives people and AI a shared work packet, policy boundary, and delivery record—while every contributor keeps the tools that make them effective.Matrix 让人与 AI 共享任务包、策略边界与交付记录,同时保留各自最高效的工具。
The platform separates work, intelligence, policy, and evidence so each can evolve independently without breaking the operating loop.平台把工作、智能、策略与证据分层管理,使它们能独立演进,同时保持运营闭环稳定。
Every mission carries its outcome, scope, context, acceptance gates, and write boundaries. It can move without losing what matters.
每项任务都携带目标、范围、上下文、验收门槛与写入边界,流转时不会丢失关键条件。
Leaders, specialists, AI agents, models, tools, and connected machines become routable resources with explicit availability and policy.
Leader、专家、AI Agent、模型、工具与在线机器成为可调度资源,并明确展示可用性与策略。
Identity, permission, scope, concurrency, cost, review gates, and environment policy stay attached throughout execution.
身份、权限、作用域、并发、成本、评审门禁与环境策略,在整个执行过程中始终随任务保留。
Artifacts, logs, decisions, reviews, and outcomes land in one durable ledger that can be resumed and audited later.
交付物、日志、决策、评审与结果进入同一份长期账本,之后仍可恢复、追踪与审计。
Accepted work becomes reusable context, components, patterns, and evidence—not exhaust trapped in a transient conversation.
已验收工作沉淀为可复用的上下文、组件、模式与证据,而不是困在一次性对话里的废气。
Missions, workers, capacity, approvals, cost, delivery health, and evidence completeness share one control plane.
任务、Worker、容量、审批、成本、交付健康度与证据完整性集中在同一控制平面。
Current Pool and DAG design · September 10, 2026. The workflow remains under validation.2026 年 9 月 10 日更新的 Pool 与 DAG 设计;完整流程仍在验证中。
Personal and company resources share one model. Each card identifies its owner, account quota, connection, access mode and available slots. Keep your lent-out resources visible.
个人与企业使用同一种资源卡,区分所有者、账号额度、连接、接入模式与可用槽位;左侧保留自己的出借资源列表。
Choose a slot count before confirming a drag into Standby. Paid capacity starts billing when the server accepts the lease. Return releases it; credentials and ownership stay with the provider.
拖入 Standby 前确认槽位数量;付费容量从服务端接受租赁时计费,退还后结束。拖拽不转移凭据或所有权。
Reserve a slot before launching an agent. Working follows each actual execution while unused capacity remains in Standby. Shared quota meters show their freshness or an explicit unknown state.
启动 Agent 前先原子占槽;Working 展示实际执行实例,剩余容量仍在 Standby。共享额度血条注明更新时间,未知就显示未知。
Each LLM node has its own inputs, outputs and any real Leader instructions. Feedback rounds retain history. Completion must include the task result, resource release and a subsequent run.
每个 LLM 节点保留自己的输入、输出与实际 Leader 指令,反馈轮次保留历史;完整闭环包含任务结果、资源释放和再次运行。
Validation is incomplete: repeat runs on the same Task and some serial dispatches still need repair. Feedback retry has been exercised; automatic graph expansion has not yet passed acceptance.验收尚未全部通过:同一 Task 再次运行及部分串行派发仍需修复。反馈重试已有实测,运行中自动扩图尚未通过验收。
Matrix removes the coordination tax between a goal and the resources that can deliver it.Matrix 消除目标与可交付资源之间反复协调的成本。
The brief, boundaries, history, and evidence move together.目标、边界、历史与证据始终一起流转。
Independent work reaches independent resources without losing one shared outcome.独立任务交给独立资源,同时保持同一个交付目标。
Review the work, artifacts, and evidence—not a persuasive summary.验收真实工作、交付物与证据,而不是一份有说服力的摘要。
State, responsibility, and proof stay visible. If a resource is unavailable, a policy is unmet, or evidence is missing, Matrix says so.状态、责任与证明始终可见。资源不可用、策略未满足或证据缺失时,Matrix 会明确说明。
Connected AI accounts remain on the machine that owns them. Matrix schedules declared capability instead of collecting provider credentials.
已连接的 AI 账号保留在所属机器上;Matrix 调度已声明的能力,而不是集中收集服务商凭据。
Write scopes, review gates, and worker policy are part of the work packet and the resulting ledger.
写入范围、评审门禁与 Worker 策略既属于任务包,也会进入最终账本。
Missing execution, unavailable providers, or incomplete evidence never become a fabricated success.
执行缺失、服务商不可用或证据不完整时,系统不会把它们伪装成成功。