湖北省星昊智能科技有限公司 · 自研产品 Hubei Xinghao Intelligent · Proprietary Product

CAE Agent
会干活、可审计的仿真代理 An auditable agent that runs simulation

不是聊天套壳。CAE Agent 自研调度内核：理解工程意图、驱动物理会话、 实时观测残差，并在关键节点暂停等人确认。现已推出浏览器端 CAE Agent Studio：任务收件箱一键下发，Best-of-N 方案竞赛择优， 任务差异像代码评审一样审阅后合并，让 AI 辅助下的结果可复现、可追责、符合工程纪律。

Not a chat wrapper. CAE Agent is a proprietary orchestration kernel: it understands engineering intent, drives physics sessions, watches residuals live, and pauses for human confirmation at critical gates. Now with CAE Agent Studio in the browser: submit tasks from an inbox, race Best-of-N candidates, and review task diffs like code before merging — AI-assisted results stay reproducible, accountable, and engineering-disciplined.

体验产品示意Try the Demo 预约实机演示Book a Live Demo
人机共驾Human-in-the-Loop
关键步可暂停 · 工程师最终拍板Pause at gates · Engineer decides
仿真记忆Simulation Memory
目标 · 边界条件 · 执行审计Goals · BCs · Audit log
方案竞赛Best-of-N Racing
多候选并行试探 · 打分择优Parallel candidates · Score & pick
评审合并Review & Merge
任务差异逐文件审阅 · 一键合入Per-file task diffs · One-click merge

WHY CAE AGENT

自研能力，而非套壳拼装Built for CAE — not bolted-on chat

星昊智能把「调度、记忆、人机闸门、任务隔离、方案竞赛、差异评审」做成产品内核，企业 SOP 与求解器插件可私有沉淀，形成真正可交付的仿真代理。 Xinghao built orchestration, memory, human gates, task isolation, candidate racing, and diff review into the product core. Enterprise SOPs and solver plugins stay private — a simulation agent you can actually ship.

调度内核常驻Always-on Leader

长驻 Leader 守护会话与任务队列：断线可重连、任务排队执行、历史全量归档，重启后自动续排未完成任务。 A long-lived Leader owns sessions and the task queue: clients reconnect, tasks queue up, history persists, and unfinished work re-queues after a restart.

人机闸门Human Gates

网格、边界条件、发散风险等关键节点，Agent 主动暂停等人确认，再继续执行。AI 提速，工程师掌舵。 At mesh, BC, or divergence risk, the Agent pauses for confirmation before continuing. AI accelerates; engineers steer.

仿真记忆白板Memory Whiteboard

工程目标、已确认边界条件、执行审计日志结构化落盘；任务内还可带上下文追问，对话全程留痕可回放。 Goals, confirmed BCs, and an execution audit trail persist as structured memory — with in-task follow-up conversations recorded and replayable.

任务级隔离工作区Isolated Task Workspaces

每个任务在独立分支工作区中执行，主工程树零污染；沙箱试探与方案对比敢放手跑，不怕搞坏基线。 Every task runs in its own branch workspace — the main tree stays untouched, so sandboxed probes and alternatives run freely without risking the baseline.

差异评审后合入Review Before Merge

任务完成先进「待评审」：逐文件差异审阅、增删行统计、冲突提示，确认后一键合入主线，全程可追责。 Finished tasks enter review: per-file diffs, add/delete counts, conflict detection — approve, then one-click merge to mainline with full accountability.

企业能力私有沉淀Private Enterprise Moat

行业 SOP、材料库、审批门槛与求解器插件可闭源部署在企业侧。平台是底座，know-how 仍是您的资产。 Industry SOPs, material libraries, approval gates, and solver plugins deploy privately on your side. The platform is the foundation — your know-how stays yours.

PRODUCT DEMO

产品示意：一个任务的完整旅程Product Demo: A Task's Full Journey

下面是 CAE Agent 工作台的网页示意（非实机）。点击播放，观看任务下发、隔离工作区开工、残差收敛、闸门等人，直到差异评审、一键合入主线的全过程。 A web mock of the CAE Agent workspace (not a live solver). Hit play to watch a task get submitted, run in an isolated workspace, converge, pause at a gate — then pass diff review and merge to mainline.

cae-agent · task

Agent 执行流Agent Feed

示意为产品交互预览。实机演示含企业求解器插件与私有 SOP 配置。 Interactive preview only. Live demos include your solver plugins and private SOP configuration.

CAE AGENT PLATFORM

从意图到收敛：完整代理链路From intent to convergence

CAE Agent 把自然语言任务落到可观测的物理会话上：每一步可检查、可暂停、可写入企业记忆。 CAE Agent turns natural-language tasks into observable physics sessions — every step inspectable, pausable, and writable into enterprise memory.

什么是 CAE Agent？What is CAE Agent?

传统仿真依赖工程师在 GUI 与脚本间切换：设参、提交、盯收敛、写报告。CAE Agent 用意图理解 + 工具白名单 + 人机闸门把重复链路自动化，同时把关键决策权留在工程师手里。 Traditional simulation jumps between GUIs and scripts — set parameters, submit, watch convergence, write reports. CAE Agent automates the repetitive chain with intent understanding, a tool whitelist, and human gates, while keeping critical decisions with engineers.

平台北向是自研调度与工作台，南向对接企业已有求解能力与私有插件。您买到的是可落地的代理系统，不是通用聊天机器人。 Northbound is Xinghao's orchestration and workspace; southbound connects to your existing solvers and private plugins. You get a shippable agent system — not a generic chatbot.

caeagent.com

Intent → Physics → Audit

  • 标准仿真可对话发起，资深工程师聚焦例外工况Standard cases start in conversation; seniors focus on exceptions
  • 执行过程全程审计，合规与复盘有据可查Full execution audit for compliance and post-mortems
  • 企业 SOP 与插件私有化，能力资产不外泄Private SOPs and plugins — your capability stays in-house

产品能力Product Capabilities

INTENT

工程意图解析Engineering Intent

把「支架 500N 静力、关注应力热点」映射为可执行步骤，而不是空泛闲聊。 Maps “500 N static on the bracket, watch stress hotspots” into executable steps — not empty chat.

TOOLS

白名单工具调用Whitelisted Tools

仅允许执行、检查、记忆写入、人机暂停、沙箱等受控动作，杜绝随意改文件与越权操作。 Only controlled actions — exec, inspect, memory, pause, sandbox — no arbitrary file or privilege abuse.

LOOP

执行—观测闭环Exec–Inspect Loop

每步物理执行后强制观测残差与状态，再决定下一步，贴合真实仿真纪律。 After every physics step, inspect residual and state before deciding the next move — real simulation discipline.

GATE

人机暂停 / 恢复Pause / Resume

卡住、网格存疑、策略需变更时，Agent 主动停住等确认，避免无人值守跑飞。 When stuck or mesh/strategy is doubtful, the Agent stops for confirmation — no unattended runaway jobs.

MEMORY

记忆与审计Memory & Audit

目标、边界条件与操作日志写入工程白板，支持换人接手与质量抽查。 Goals, BCs, and action logs land on the project whiteboard for handoffs and quality review.

SANDBOX

沙箱并行试探Parallel Sandboxes

多方案对比在隔离副本中进行，选定后再合入主线，降低试错成本。 Compare alternatives in isolated copies, then merge the winner — lower cost of being wrong.

BEST-OF-N

方案竞赛择优Candidate Racing

N 个候选各立假设、各自探索，评分板排序，晋升前可预览合入内容。 N candidates state hypotheses and explore; a scoreboard ranks them, with a merge preview before promotion.

REVIEW

差异评审合并Diff Review & Merge

任务改动逐文件审阅，冲突明确提示，确认后一键合入主线。 Review task changes file by file, surface conflicts explicitly, then merge to mainline in one click.

TEAM

角色分权协同Role-Based Teamwork

工程师、专家、负责人、管理员各见其所需，权限矩阵服务端强制执行。 Engineers, experts, managers, and admins each see what they need — the permission matrix is enforced server-side.

典型闭环Typical Loop

  1. 1
    下达任务Submit Task

    在收件箱描述工况与验收标准，选择委托 / 共驾 / 方案竞赛模式。 Describe conditions and acceptance criteria in the inbox; pick delegate, copilot, or Best-of-N mode.

  2. 2
    隔离开工Isolate & Start

    任务领取独立分支工作区，拉起物理会话，主工程树不受影响。 The task gets its own branch workspace and opens a physics session — mainline stays untouched.

  3. 3
    步进与观测Step & Inspect

    小步执行 → 读残差 → 再决策；右栏曲线同步更新。 Small exec → read residual → decide again; the residual pane updates live.

  4. 4
    闸门确认Gate Confirm

    遇风险暂停；工程师确认后恢复，全程记入审计。 Pause on risk; engineer confirms; resume — all logged to the audit trail.

  5. 5
    评审合入Review & Merge

    达标后进入待评审：逐文件审阅差异，确认合入主线并归档复用。 On acceptance the task enters review: inspect per-file diffs, merge to mainline, archive for reuse.

落地场景Where It Lands

标准件批跑Standard Batch Runs

重复工况交给 Agent 闭环执行，资深工程师只在闸门处介入。 Hand repetitive cases to the Agent loop; seniors intervene only at gates.

方案竞赛Candidate Racing

网格密度、关闭规律等多方案 Best-of-N 并行竞赛，评分择优后评审合入。 Race mesh densities or closure laws Best-of-N style; score, review, then merge the winner.

交接与合规Handoff & Compliance

新同事打开记忆白板即可看懂「目标—BC—做过什么」，审计可抽查。 New teammates read the whiteboard: goal, BCs, what ran — audits become practical.

跨团队入口Cross-Team Entry

设计 / 工艺用统一意图描述发起标准仿真，减少排期扯皮。 Design and process teams launch standard sims from shared intent — less scheduling friction.

支持私有化部署与企业插件定制。预约实机演示，看您的 SOP 如何变成可执行代理。 On-premises deployment and custom enterprise plugins. Book a live demo to see your SOP become an executable agent.

预约 CAE Agent 演示Book a CAE Agent Demo

NEW · CAE AGENT STUDIO

Studio：浏览器里的仿真任务工作台 Studio: The Simulation Workspace in Your Browser

最新发布。像管理代码评审一样管理仿真任务：收件箱下发、闸门审批、方案竞赛、差异评审、运维监控——团队各角色在同一工作台协同，无需安装客户端。 Just shipped. Manage simulation like code review: submit from an inbox, approve gates, race candidates, review diffs, and monitor operations — every role collaborates in one workspace, no client install.

仿真工程师Engineer CAE 专家Expert 项目负责人Manager 平台管理员Admin 令牌登录 · 角色分权 RBAC · 事件实时推送 Token login · Role-based access · Live event streaming

自然语言下发，全生命周期在管Submit in plain language, managed end to end

  • 三种执行模式：全权委托 / 人机共驾 / Best-of-N 方案竞赛Three modes: delegate, copilot, or Best-of-N racing
  • 排队 → 执行 → 闸门 → 评审 → 完成，状态实时推送Queued → running → gate → review → done, streamed live
  • 闸门审批与任务终止直接在收件箱完成Approve gates or abort right from the inbox
  • 任务历史持久归档，服务重启不丢；任务内可带上下文追问History survives restarts; ask follow-ups with full task context

人与 Agent 轮流驾驶同一会话You and the Agent share one session

  • 会话树、记忆白板、事件流三栏同屏，Agent 思考过程实时可见Session tree, whiteboard, and feed side by side — thoughts stream live
  • 对数残差曲线带阈值线，收敛趋势一眼可判Log-scale residual with threshold line — convergence at a glance
  • 求解器画面截图直达浏览器，无需远程桌面Solver screenshots in the browser, no remote desktop
  • 随时插入指令接管，或把方向盘还给 AgentType to take over anytime, then hand the wheel back

让 N 个方案先赛一场，赢家才进主线Race N candidates; only the winner lands

  • 一键派生 N 个候选沙箱，Agent 各自设定假设并探索Spawn N sandboxes at once; each Agent states and tests a hypothesis
  • 评分板按残差自动排序，探索过程全程可查A scoreboard ranks by residual; every run stays inspectable
  • 晋升前逐文件预览将合入的内容，专家确认后才生效Preview exactly which files merge back — nothing lands without expert sign-off
  • 失败方案自动记录经验，沉淀为下次探索的先验Failed candidates leave lessons behind as priors for the next search

仿真改动，像代码一样评审Review simulation changes like code

  • 每个任务在独立分支工作区执行，主工程树零污染Each task runs on its own branch workspace — mainline untouched
  • 完成后自动进入待评审：逐文件差异、增删统计、二进制识别Finished tasks queue for review: per-file diffs, counts, binary detection
  • 确认后一键合入；有冲突则明确列出，主线绝不半合Approve to squash-merge; conflicts are listed — mainline never half-merges
  • 合并记录含任务号与操作人，审计链闭环Merges record task id and actor — the audit chain stays closed

给负责人和管理员的全局视角The bird's-eye view for managers and admins

  • 运行统计、逐日批跑与闸门响应时长，明细可导出 CSVRun stats, daily batches, gate-response times — export details to CSV
  • 白板审计时间线，谁批准了什么一目了然A whiteboard audit timeline shows who approved what
  • 节点健康、插件注册表、许可证池与 RBAC 矩阵实时可查Node health, plugin registry, license pool, and the live RBAC matrix
  • LLM 配置脱敏展示，密钥不出服务器LLM config shown masked — keys never leave the server

交付形态：网关统一鉴权托管，浏览器即开即用；Windows 安装包（zip / MSI）配套 CI 流水线；TUI 与命令行自动化接口同级支持，脚本化批跑无缝衔接。 Delivery: gateway-hosted with unified auth — open a browser and go; Windows zip/MSI packages with a CI pipeline; TUI and CLI automation stay first-class for scripted batch runs.

FLAGSHIP DELIVERY

标杆交付 · 水力过渡过程智能计算平台 Flagship · Hydraulic Transient Platform

面向抽水蓄能电站输水系统，以特征线法（MOC）为计算内核，打通「一维模型自动建立 → 瞬态过渡过程求解 → 极值包络可视化」的完整工作流。这是星昊把仿真能力真正交付进工程流水线的行业标杆。 For pumped-storage waterway systems: a method-of-characteristics (MOC) kernel spanning the full workflow — 1D model generation, transient solving, and envelope visualization. A benchmark of how Xinghao ships simulation into the engineering pipeline.

7 类核心组件全覆盖Core component types
50+ 六大类典型工况Typical operating cases
11,708 步瞬态迭代实测Transient steps run
MOC 特征线法求解内核Characteristics kernel

三大核心验证案例Three Validation Cases

案例 01 · 经典水击 Case 01 · Water Hammer

单管-阀系统 · Joukowsky 效应再现 Pipe–valve system · Joukowsky effect

阀门瞬间关闭，动能骤然转化为压差能。平台精确捕捉压力「方波」周期震荡，再现极限工况下的能量守恒与应力波传播、反射机制。 An instant valve closure converts kinetic energy into pressure. The solver captures the square-wave pressure oscillation, reproducing energy conservation and wave propagation/reflection.

  • 水击升压 1176.66 mSurge 1176.66 m
  • Joukowsky 理论 1081.32 mJoukowsky theory 1081.32 m
  • 波速 1200 m/s · 稳态流速 8.84 m/sWave 1200 m/s · V₀ 8.84 m/s
案例 02 · 电站甩负荷 Case 02 · Load Rejection

双机组抽水蓄能电站 · 满负荷甩负荷 Dual-unit station · full load rejection

两台机组同时解列，工况点穿越复杂「S」区。3960 步瞬态迭代平滑输出转速飞逸曲线，调压井长周期涌波有效保护上游隧洞。 Two units trip together, crossing the tricky “S” zone. 3,960 transient steps yield a smooth runaway curve; surge-tank oscillation protects the upstream tunnel.

  • 最大转速飞逸 540.0 rpmRunaway 540.0 rpm
  • 调压井最高涌波 622.69 mSurge peak 622.69 m
  • 支管最大水头 845.6 mBranch head 845.6 m
案例 03 · 全要素工程 Case 03 · Full System

三机组综合工程 · 450MW 满发甩负荷 Three-unit 450 MW · full system

单一算例覆盖全部七类组件，同时呈现毫秒级水击波与百秒级质量振荡两种时间尺度。11708 步迭代，全部关键指标落于设计限值内。 One case covers all seven component types, spanning millisecond water-hammer and hundred-second mass oscillation. 11,708 steps; every key metric within design limits.

  • 支管最大水头 996.8 m（限 1083 m）Branch 996.8 m (limit 1083 m)
  • 机组转速 488.2 rpm（限 540 rpm）Speed 488.2 rpm (limit 540 rpm)
  • 上下游调压井涌浪均达标Both surge tanks within range

平台核心能力Platform Capabilities

一维模型自动建立Auto 1D Modeling 从 Excel / CAD 识别节点与管段，拖拽式拓扑编辑，输出平面图与剖面图。 Parse nodes and pipes from Excel/CAD, drag-to-edit topology, output plan and profile views.
MOC 瞬态求解内核MOC Transient Solver 一维非恒定有压管流求解，多管段统一时间步长，兼顾稳定性与精度。 1D unsteady pressurized-flow solving with unified time-stepping for stability and accuracy.
全特性曲线 Suter 转换Suter Full-Characteristics 水泵水轮机四象限全特性处理，攻克甩负荷、断电、飞逸等极端工况收敛。 Four-quadrant pump-turbine characteristics; convergence under rejection, blackout, and runaway.
多类边界条件Rich Boundary Conditions 调压室 / 闸门井、球阀、岔管、水库等边界统一联立求解。 Surge tanks, ball valves, bifurcations, and reservoirs solved in one unified system.
极值包络线可视化Envelope Visualization 自动遍历全流道极值，生成沿程压力包络线，支持多工况叠加对比。 Auto-scan the flow path for extremes and plot pressure envelopes with multi-case overlay.
历时曲线与工况管理Time Series & Case Mgmt 观测节点多参数历时曲线，六大类工况参数化定义与批量并行计算。 Multi-parameter time histories, six operating-condition families, and batch parallel runs.

需要为抽水蓄能、水电或供水系统定制水力过渡过程计算平台？我们可基于该内核做私有化交付。 Need a hydraulic-transient platform for pumped-storage, hydropower, or water supply? We deliver privately on this kernel.

预约平台演示Book a Platform Demo

INDUSTRY SOLUTIONS

行业解决方案Industry Solutions

以 CAE Agent 为底座，按行业沉淀私有流程与验收判据。 CAE Agent as the base — private workflows and acceptance criteria by industry.

AUTOMOTIVE

汽车与交通装备Automotive & Transportation

结构、NVH、热管理等标准分析流程 Agent 化；关键安全判据保留人机闸门。 Agentize structural, NVH, and thermal standard flows — keep human gates on safety criteria.

ENERGY

能源与动力设备Energy & Power Equipment

多工况批跑与运维复盘结合审计记忆，让可靠性分析可追溯。 Multi-condition batch runs plus audit memory make reliability analysis traceable.

MANUFACTURING

先进制造与装备Advanced Manufacturing

工艺仿真与强度评估模板化，设计—仿真—现场反馈闭环可编排。 Templated process and strength evaluation; orchestrate design–sim–field feedback loops.

ENTERPRISE

企业私有 Agent 套件Private Agent Suites

按贵司规范预置闸门、材料与报告口径；插件与 SOP 不出企业边界。 Pre-wire gates, materials, and report formats to your standards — plugins and SOPs stay inside your perimeter.

ABOUT US

关于星昊智能About Xinghao Intelligent

湖北省星昊智能科技有限公司 深耕工程仿真与工业智能。我们不做「大模型 + 说明书」的演示玩具，而是交付能进入研发流水线的仿真代理系统。 Hubei Xinghao Intelligent Technology Co., Ltd. focuses on engineering simulation and industrial intelligence. We don't ship “LLM + docs” demos — we deliver simulation agents that enter the R&D pipeline.

产品主线 CAE Agent（caeagent.com）强调自研调度、人机共驾与可审计记忆。企业求解能力与 SOP 以私有插件方式接入，形成可持续迭代的研发基础设施。 Our product line CAE Agent (caeagent.com) emphasizes proprietary orchestration, human-in-the-loop control, and auditable memory. Enterprise solvers and SOPs plug in privately as lasting R&D infrastructure.

  • 工程纪律：可暂停、可观测、可复现Engineering discipline: pausable, observable, reproducible
  • 能力归属：企业 know-how 私有沉淀Ownership: enterprise know-how stays private
  • 交付路径：PoC → 私有化 → 规模化批跑Delivery path: PoC → on-prem → scaled batch

CONTACT

开启合作Start Collaboration

告诉我们您的仿真场景与落地约束，我们将在 1–2 个工作日内回复。 Tell us your simulation scenarios and deployment constraints. We reply within 1–2 business days.

公司全称Company 湖北省星昊智能科技有限公司Hubei Xinghao Intelligent Technology Co., Ltd.
官网Website caeagent.com
邮箱Email contact@caeagent.com
所在地Location 湖北省Hubei Province, China