Lyapunov Alpha8

Lyapunov Alpha8

Alpha8（0.1.0-alpha.8）是预发布版本，继续固定现有RC2 SDK。本版修复世界关闭失败后的重试，并将已核验的原生物理状态、材料参数与桌面保存重开工作流接通。

• 修复 Issue #1：关闭失败后保留世界并允许重试，成功后清理世界与附加缓存；同一世界的并发关闭共享一次在途请求。
• MuJoCo接通原生flex／实体cable的材料、形变与接触。Newton接通受限源位置控制、首帧暂停、原生RGB-D相机与静态容器water MPM。Isaac接通已有USD可变形体／PBD粒子的参数及原生节点／粒子读回。
• 原生参数面板和 scene_physics_update 的 native 入参分支共用局部材料更新；同步后可读回实际值，再另存与重开。
• 同名工程可用“打开副本 / Open copy”创建新场景，保留现有场景及资源。
• 自动检查点限定所选工程目录，准备过程支持取消，保留既有恢复数据。

源码：99372b88c521878a17f3e96f52f4845600c63987。公开仓库本次准确CI。Ubuntu26.04.1正常安装与默认Mu工作流结果：正常安装退出0；同进程“打开副本”生成新场景身份，默认Mu重新加载后实际读回Young模量40000 Pa、泊松比0.3、阻尼0.001与27顶点，正常退出；原示例及XML哈希保持。

版本化下载

• Linux x64主包：739914751字节，SHA256 67b7e661b35c5b6f49b7f6ed08b8ce2baf1bc25abef22f2e55e173758895fbc3。
• 默认MuJoCo CPU runtime：172664865字节，SHA256 b8e090b9845a646ea7457f26f726757e1d40784603b55fe7ff2058cb19e033cb。
• SHA256SUMS、完整JSON清单、TSV清单。GitHub同时附这三件小文件；归档由既有OSS/CDN下载链提供。

本版实现证据限于已验的MuJoCo flex／cable、Newton受限控制／静态容器MPM／原生相机及Isaac两块8节点软硬体与64粒子的GPU案例。Isaac需要有效的本机SDK和GPU；不要求三引擎同时启动。动态机器人／MPM耦合、逐粒子碰撞力、任意外部USD依赖、磁场／磁流体和更广API留待后续。视觉设计与RC留下一阶段。

English

Alpha8 (0.1.0-alpha.8) is a prerelease that retains the existing pinned RC2 SDK. It fixes retries after world-close failures and connects verified native physics state and material parameters to the desktop save-and-reopen workflow.

• Fix Issue #1: a failed close retains the world and allows a retry; successful closure removes the world and related caches. Concurrent closes of the same world share one in-flight request.
• MuJoCo connects native flex and physical cable materials, deformation and contacts. Newton connects restricted source-position control, initial-frame pause, native RGB-D cameras and static-container water MPM. Isaac connects parameters and native node or particle observations for existing USD deformables and PBD particles.
• The native-parameter panel and the native input branch of scene_physics_update share local material updates. Synchronize, read actual values, then save a copy and reopen it.
• Use Open copy for a project with the same scene identity to create a fresh scene while preserving the existing scene and its resources.
• Automatic checkpoints are scoped to the selected project directory. Preparation can be cancelled while existing recovery data is retained.

Source: 99372b88c521878a17f3e96f52f4845600c63987. Exact public-repository CI. Normal Ubuntu 26.04.1 installation and default Mu workflow result: installer exit 0; same-process Open copy created a fresh scene identity, and the default Mu runtime reloaded it with actual readback of Young modulus 40000 Pa, Poisson ratio 0.3, damping 0.001 and 27 vertices; normal exit, with original demo and XML hashes preserved.

Versioned downloads

• Linux x64 main archive: 739914751 bytes; SHA256 67b7e661b35c5b6f49b7f6ed08b8ce2baf1bc25abef22f2e55e173758895fbc3.
• Default MuJoCo CPU runtime: 172664865 bytes; SHA256 b8e090b9845a646ea7457f26f726757e1d40784603b55fe7ff2058cb19e033cb.
• SHA256SUMS, complete JSON manifest, and TSV manifest. These three small files are also GitHub assets; archives use the existing OSS/CDN download channel.

Implementation evidence is limited to the verified MuJoCo flex/cable, restricted Newton control, static-container MPM and native cameras, and the Isaac GPU case with two eight-node soft/stiff bodies and 64 particles. Isaac requires a valid local SDK and GPU; all three engines need not run together. Dynamic robot/MPM coupling, per-particle collision forces, arbitrary external USD dependencies, magnetic fields or magnetohydrodynamics, and broader APIs remain future work. Visual design and RC work belong to the next phase.

默认安装入口

默认安装器、latest清单和该版本两条归档下载路线已切换并读回。Linux x64继续使用正常默认MuJoCo CPU runtime与桌面入口：

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh -s -- --version 0.1.0-alpha.8

既有版本、previous与用户数据按原安装器合同保留。使用方式与依赖条件见下载入口和完整清单。

Default installation entry

The default installer, latest manifest and both versioned archive routes have been switched and read back. Linux x64 retains the normal default MuJoCo CPU runtime and desktop entry:

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh -s -- --version 0.1.0-alpha.8

Existing versions, previous and user data follow the installer’s existing retention contract. See the download entry and complete manifest for usage and dependencies.

Lyapunov Alpha7

Lyapunov Alpha7

Alpha7 将电脑操作、Blender 与 Unity 的原生 MCP 接入工作台，保留已有安装与用户配置。设置页整合软件连接、Agent 预设、物理、机器人库和可选下载；3D 工作台提供中键导航、所选资产加入聊天的文件引用与表面批注。默认机器人库包含九个许可资产闭包已核对的模型，并供给 Go1/G1 基础控制策略。光照与材质新增可保存的色调映射、HDRI 方向、阴影、面积灯和实体独立 PBR 控制，并保留高精场景首相机与坐标修复。

Linux 支持 Ubuntu 24.04 或更新版本、glibc 2.39 或更新版本。Unity 编辑器与 addon 仍需正常安装和连接；缺少许可证的 Go2 策略不随包分发，其余未适配控制项明确显示缺项。正常安装、升级和回滚保留用户数据。安装与下载使用本版本正式清单。Wayland 键盘兼容限制见 README 中文说明。

English

Alpha7 brings native computer control, Blender and Unity MCP into the workbench while preserving existing installations and user configuration. Settings bring together software connections, agent presets, physics, the robot library and optional downloads. The 3D workbench adds middle-button navigation, asset references in chat and surface annotations. The default library includes nine models with verified licensed asset closures and basic-control policies for Go1 and G1. Persistent rendering controls cover tone mapping, HDRI orientation, shadows, area lights and independent per-entity PBR materials, while preserving the high-detail scene first-camera and coordinate fixes.

Linux requires Ubuntu 24.04 or newer and glibc 2.39 or newer. Unity's editor and addon require normal installation and connection. The unlicensed pinned Go2 policy is not distributed, and unsupported controls are identified explicitly. Normal installation, upgrade and rollback preserve user data. Installation and downloads use this release's final manifest. See the README's Wayland keyboard compatibility note.

安装与校验 / Installation and verification: 官网安装页 / Installation page · Alpha7 正式清单 / Final manifest · 校验和 / Checksums。

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh -s -- --version 0.1.0-alpha.7

遇到问题请联系 voryneltech@gmail.com。For help, contact voryneltech@gmail.com.

© 2026 Vorynel Co., Ltd. · Apache License 2.0 · 赞助方 / Sponsor: UAD（浙江大学建筑设计研究院有限公司）。

Lyapunov Alpha 6.1.1

修复自动生成的官方高精场景在客户端首次打开时的默认朝向、机位和画质，并统一模型放置时的范围计算。文件预览和工作台使用同一套场景识别规则。

验证：主控完整测试342/342，两套类型检查、插件和桌面构建、资源布局检查通过；同一源提交的GitHub Actions已通过。旧版合格场景与新版生成场景均完成正式包默认导入、硬件渲染和模型文件一致性测试。

官网安装与下载 · 版本清单

程序来源：53379f6a9986de7674edbc3d209ed80fc7e29dfe。Linux x64主程序SHA-256：0864c1b3cb04c1760efc2fc71f99735ca82c807e3bc6ccad5939ebcd926510a1。

Lyapunov 0.1.0 Alpha 6.1

Lyapunov 0.1.0-alpha.6.1

English README · 中文 README · Installation / 安装

English

This Linux x64 Alpha keeps the owned DSH fork pinned to 0.2.0-rc.2 and improves native robot assembly and position-control declarations.

• Fixed-base attachments use native parent-body assembly. A dexterous hand attached to an arm follows its selected link without relying on a soft weld. Missing targets, self-targets and cyclic attachments are rejected.
• Pure-torque motor models require explicit position-loop gains before position commands. Importing G1 does not manufacture a standing or walking controller; load a compatible, licensed policy or declare the intended controller before motion.
• Unanchored fixed-base models can be lifted out of an initial ground overlap with a recorded warning. Explicit world or body anchors retain their requested positions.
• The application version shown on the welcome page now comes from the same frozen version as the release manifest. The installer keeps the previous version and user data.

The package includes the managed MuJoCo runtime. Isaac remains optional and requires its own supported runtime, driver and GPU. Large environment collision derivation may take time and memory; a visible mesh alone does not prove that its collision or physics is ready.

For direct support, email voryneltech@gmail.com. Report reproducible problems in Issues and share ideas in Discussions.

中文

此 Linux x64 Alpha 继续锁定自有 DSH fork 的 0.2.0-rc.2，改进原生机器人装配与位置控制声明。

• 固定基座使用原生父连杆装配。安装在机械臂上的灵巧手跟随指定连杆，不依赖软焊接。缺失目标、自身目标和循环挂载会明确拒绝。
• 纯力矩电机模型执行位置指令前必须显式声明位置环增益。导入 G1 不会凭空生成站立或行走控制器；运动前需加载适配且获许可的策略，或声明所需控制器。
• 未显式锚定的固定基座模型可从初始地面重叠中抬升，并记录警告。显式世界或连杆锚点保留请求的位置。
• 欢迎页显示的应用版本与发行清单采用同一份冻结版本。安装器保留上一版本和用户数据。

安装包包含独立管理的 MuJoCo 运行环境。Isaac 为可选功能，需要受支持的独立运行环境、驱动和 GPU。大型环境的碰撞派生可能需要较多时间和内存；能够看到网格，并不代表碰撞和物理已就绪。

如需直接协助，请邮件联系 voryneltech@gmail.com；可复现问题请提交至 Issues，方向与建议欢迎在 Discussions 讨论。

Downloads / 下载

Install / 安装:

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

Linux x64 package / 主安装包 · MuJoCo runtime / 运行包

The final package was installed and opened on Ubuntu2604 under QEMU/KVM. Native robot control, mounted camera persistence and the infinite ground were verified. Windows/WSLg and RTX execution are outside this acceptance.

最终安装包已在 QEMU/KVM 的 Ubuntu2604 上实际安装并打开，验收了原生机器人控制、挂载相机持久保存和无限地面。本次验收不代表 Windows/WSLg 或 RTX 已通过。

Lyapunov 0.1.0-alpha.5

Lyapunov 0.1.0-alpha.5

Lyapunov is an LLM and native 3D first robotics framework produced by Vorynel Co., Ltd.（杭州奇异宇宙人工智能有限公司）, with support from UAD（浙江大学建筑设计研究院有限公司）. It builds on DeepSeek Harness, MuJoCo, Isaac Sim, and their upstream communities.

Install

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

The Linux x64 installer supplies private Node, Electron, and DSH runtimes and prepares the MuJoCo companion. Downloads use OSS/CDN, support resumption, and verify manifest sizes and SHA-256 values. Isaac remains an optional, explicitly licensed installation.

Changes

• Upgrade the DSH base to official 0.2.0 RC2, integrated and preserved in the project's own fork. Pin commit 54312dc5a8c4a63b60b3b92f51109f7bd1c1e852 and retain the version tag for reproducible future checkout.
• Adapt native configuration, Session V4 history, references, and Jobs consumers while preserving Lyapunov's Agent Graph scheduling and native 3D context.
• Correct portable native-loader dependency resolution, RC2 terminal preset imports, and clean-checkout history test paths.
• Preserve Alpha4 robot, policy, camera, managed ground, local file, language, and installation features.

Validation and known limits

Source: e3030061e16fc54c65eba66e4ad541f12dc50f34. Dev CI and Issue CI passed for this exact commit. The default offline set contains 339 passing entries; environment-specific entries remain separate.

Sandboxed package Host/API checks exercised MuJoCo and Isaac CPU/PhysX small worlds, native gravity and contacts, paused initialization, save/reopen, invisible ground support, synchronized ground deletion and collision-disable negatives, and Stop/close. Ten of the eleven current T0 models completed bounded native import and control under their stated installation conditions. These checks reuse already installed private SDKs and do not establish a new VM, GPU/RTX, cold install, learned policy, or every robot interaction.

This is an alpha. The original Generic Quadrotor DAE mesh remains unsupported. The Panda/LEAP loaded wrist attachment exceeds the strict pose-drift tolerance. G1's original MuJoCo torque-actuator model does not establish position tracking or free standing; a compatible controller/policy is required. Learned G1 policy execution remains unverified. The complete B1 environment exceeded its scoped preparation budget; a controlled small-room case passed separately. UR10e's initial contact showed penetration and an abnormal force, and is not accepted as valid interaction. These issues are retained for the follow-up repair.

Downloads and documentation

• Linux x64 desktop package
• MuJoCo companion
• Checksums
• 中文使用文档
• English README · 中文 README

Apache-2.0 applies to project-owned code. Upstream copyright and licenses remain applicable. Discussions and corrective contributions are welcome.

---

中文说明

Alpha5 将底座提升到官方 DSH 0.2.0 RC2，并把完整集成版本保存在自有 fork 的固定提交与标签。Agent Graph 的核心调度逻辑保留，原生3D、机器人、相机、地面和本地工作流继续作为产品核心。

Dev与Issue精确提交CI通过；实际包的两个引擎小世界、地面与接触正负对照通过。机器人与大型B1的未完成项在上方如实列明，基本关节可动不等于站立、步态、飞行或抓取策略已完成。此次先发布，再进行独立分支修复尝试，后续修复不会覆写这一不可变发行包。

Lyapunov 0.1.0-alpha.4 — Linux x64

Lyapunov 0.1.0-alpha.4

An alpha release of Lyapunov, an LLM and native 3D first robotics framework by Vorynel Co., Ltd. Sponsored by UAD. Built on DeepSeek Harness, MuJoCo, Isaac Sim, and the work of their upstream communities.

Install

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

The Linux x64 installer includes the private Node/DSH runtime and installs the verified MuJoCo companion runtime. Package downloads use OSS/CDN, support interrupted-download resumption, and verify the release manifest.

Isaac Sim is optional. Use Settings to create a product-private Conda environment or discover and explicitly select a compatible existing Isaac Sim 6.0.1 / Python 3.12 installation. Installation requires acceptance of NVIDIA's license. The required Kit component dependency closure is included in the installation plan; the SDK is not bundled into the default desktop download.

Changes

• Repair desktop workbench activation and login recovery; preserve normal Chromium sandboxing.
• Bundle Chinese fonts and align desktop/workbench language handling.
• Restore portable native dependencies, local robot imports, directory handling and input acknowledgement.
• Correct local policy registration and reopening through private-safe entry IDs.
• Support managed infinite ground and robot-mounted camera authoring.
• Correct Isaac resource-origin versus native-root transforms and live native-root feedback.
• Repair large static environment triangle compilation, bounded work arenas and the MuJoCo 3.13 static-contact solver compatibility path.
• Use client network tools and official robot sources instead of a default model-provider web-search dependency.

Validation and alpha limits

Source: 71b9da0fc69decd51c189ff931ca637f8e3430b4. Both Dev and Issue CI passed for this exact commit. Normal sandboxed Isaac CPU/PhysX contact and G1 native control have been exercised with the product's private SDK. Ten of the eleven current T0 models passed bounded native control under the documented setup conditions; forklift is no longer a T0 target.

This is an alpha. Generic Quadrotor's DAE mesh import remains unsupported; the rigid wrist attachment demo retains a soft-weld load offset; learned standing/walking/grasping policies and GPU/RTX are not established by basic joint tests. The complete B1 collision world is a large-asset case requiring about 12 GiB for the tested process tree, before broader application/OS overhead. Its paused first frame and static contact checks are verified; the full dynamic staircase and disable-collision checks remain pending. VM installation/startup verification follows publication in this rollout.

Downloads and documentation

• Linux x64 desktop package
• MuJoCo companion runtime
• Checksums
• 中文使用文档
• English README · 中文 README

Licensed under Apache-2.0. Upstream licenses and copyright notices remain applicable to their components and model assets.

---

中文说明

Lyapunov 由杭州奇异宇宙人工智能有限公司出品，UAD 赞助，将 LLM 与原生 3D 工作流置于产品优先位置。本次 Alpha4 包含安装、工作台、字体、本体导入、策略登记、相机与地面，以及 Isaac 原点／实时反馈和大型静态碰撞修复。

默认安装基础程序和独立 MuJoCo；Isaac 在设置中按需安装，自动建立本包专用 Conda 环境，也可明确选择兼容外置 SDK。不会覆盖系统 Python 或将两套引擎依赖混装。

请先按实际机座／腕部安装本体，再保持世界暂停并匹配控制器。基本关节可动不代表已完成站立、步态、飞行或抓取策略。当前未完成项目见上方 alpha limits；欢迎提交 Discussions 和修正代码，与我们及上游社区共同完善。

Lyapunov 0.1.0α (Alpha3_5) — source prerelease

Lyapunov 0.1.0α (Alpha3_5) publishes the Linux x64 Alpha source under Apache License 2.0.

Source release identity	Value
Public name	0.1.0α (Alpha3_5)
releaseId	0.1.0-alpha3_5
Package version (SemVer)	0.1.0-alpha.3.5
Source tag	v0.1.0-alpha3_5
Source commit	5875bbe0db2d72e4de5312ba39cc59343d72af1c

The source includes the desktop workbench, scene and robot tools, camera workflows, and supported simulation integrations. See the release notes and capability matrix for this Alpha's scope.

The exact source commit passed the build, TypeScript, offline behavior, asset layout, upstream lock, and distribution artifact checks before publication.

This is a source prerelease. The website installer and downloads continue to use the batch identified by the website's current release manifest. This entry does not publish a new Linux application archive or MuJoCo companion. Previously published archives retain their original bytes, URLs, build provenance, and bundled license notices.

Produced by: Vorynel Co., Ltd. (杭州奇异宇宙人工智能有限公司). Sponsored by: UAD — The Architectural Design & Research Institute of Zhejiang University Co., Ltd. (浙江大学建筑设计研究院有限公司). We thank UAD for the support that made the development of Lyapunov possible.

Authors: Mingjun Cheng; Zongjian Ding; Yudong Gao; Yi Yang; Lidong Chen; Jiale Liu; Xinling Yu.

© 2026 Vorynel Co., Ltd. The root LICENSE contains Apache License 2.0; NOTICE retains upstream copyright and license notices.

0.1.0α（Alpha3）

Lyapunov 0.1.0α (Alpha3) — Linux x64 Alpha

“To see a World in a Grain of Sand” — William Blake, Auguries of Innocence

Lyapunov is an LLM-first, 3D-native robotics and agent desktop framework, produced by Vorynel Co., Ltd.. It brings natural-language tools, local files, scene editing, robot preparation, controlled physics and camera observations into one workspace.

English README · 简体中文 · 中文安装与产品手册 · Download and system requirements

Install

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

Public release name: 0.1.0α. Technical version and release ID: 0.1.0-alpha.3; Git tag: v0.1.0-alpha.3. The default installer selects the version from the published manifest. The Linux x64 application includes Node.js, Electron, pinned DSH and product plugins; the default MuJoCo companion uses a private, version-specific Python environment. The installer verifies published archives, relocates the companion at its final path, checks dependencies and native physics, then switches current. Failed installation retains the previous version and user data. When needed, the installer requests normal operating-system authorization, prepares the missing desktop libraries or sandbox setup, and continues installation automatically; no additional manual setup command is required.

After installation reports readiness, start Lyapunov yourself from its desktop entry or a terminal:

lyapunov

If the user launcher directory is not on PATH, use "$HOME/.local/bin/lyapunov". The installer does not automatically open the application.

• Application archive: 553,372,835 bytes; SHA-256 1e6afcbff478d26e10b53a0310a858304afc83ec56ce1a38ef92b974144a1dda.
• MuJoCo companion: 172,664,865 bytes; SHA-256 b8e090b9845a646ea7457f26f726757e1d40784603b55fe7ff2058cb19e033cb.
• Release manifest.

The versioned manifest records the actual source commit, archive sizes and SHA-256 values. The installer retains partial downloads and resumes them, then checks the actual bytes against that manifest. Previously published download URLs and objects remain unchanged.

What changed

Alpha 3 fixes the missing frontend payload behind the blank workbench seen after the welcome page in the Ubuntu 26.04 VM. The Alpha 2 archive omitted the native DSH web entry and static assets; the guest Host returned 404 or empty content. The standard packager now builds the pinned native frontend with Vite before dependency closure and includes its entry page, module JavaScript and CSS. Missing required frontend files stop packaging.

The installation and workbench fixes from Alpha 2 are retained:

• Graphical-library and sandbox preparation follows the actual desktop checks; the installer requests normal system authorization when needed and continues automatically after preparation.
• Existing desktop-entry information is kept for recovery instead of losing the previous entry during installation or repair.
• File-location displays use absolute paths and reuse the native file-browsing workflow.
• The mistakenly attached execution-graph button is removed from the default workbench UI.

The established robot, camera and local-workspace behavior is retained:

• Robot and world preparation use paused initialization followed by explicit, bounded execution. A source address or loaded SDK is distinct from verified control readiness.
• Saved camera views have a visible return to the main view. Camera binding, lens parameters, robot link frames and observations share native scene state.
• Static environment collision derivation preserves interior spaces with surface boxes; dynamic and explicit convex-decomposition paths retain their own behavior.
• Guest mode opens local features without product-account, model-relay or billing access. Agent use requires an explicitly configured personal provider.
• Robot catalog responses provide official source roots and metadata. The client uses browser, WebFetch and local tools to retrieve and check actual files; server metadata does not assert robot behavior.
• Continuous manual action feedback is consolidated, and scene/world selection, TCP, anchor and joint-control preparation remain visible in the workbench.

Documentation and examples

The English and Chinese READMEs cover environment and object construction methods, 19 input routes, robot import/download/compatibility, camera mounting and calibration, annotation editing, persistence and export. Four tutorials cover a ground/box scene, Panda joint/TCP controls, a mounted camera with annotations, and indoor collision alignment.

Native macOS, mobile/WeChat control, Genesis, VR/XR and companion DSH upgrades are listed in the roadmap. Windows uses the WSL2 + WSLg Linux-package route; customer WSLg acceptance remains pending, and this release does not claim complete Ubuntu 26 coverage. Native tools and APIs are recommended for routine workflows; CUA is not the default task interface.

Validation and Alpha scope

Source: 0fb716f93039169a6f09183d89c920e7efed8907. The standard package is built from this source and includes the pinned native DSH web frontend. Its 92 packaged frontend files were checked byte for byte against the build output. Use the archive identities recorded by the Alpha 3 manifest for this package.

This Alpha does not promise every GPU/distribution combination, full B1 navigation, arbitrary multi-robot collaboration, or G1 recovery from a fallen pose. Gait and get-up policies are separate capabilities. OS sandbox requirements are still checked by the installer; a successful desktop round is not universal platform acceptance.

Build together

Please use Discussions for ideas, Issues for reproducible problems, and pull requests for fixes. See CONTRIBUTING.

Thanks to DSH/Cordis, MuJoCo, NVIDIA Isaac Sim/IsaacLab, Three.js, Electron, Node.js, React, Blender, robot manufacturers, official model libraries and contributors. Their work and licenses are retained; Lyapunov contributes the integrated product workflow.

Authors and copyright

Mingjun Cheng; Zongjian Ding; Yudong Gao; Yi Yang; Lidong Chen; Jiale Liu; Xinling Yu.

© 2026 Vorynel Co., Ltd. The Chinese producer remains 杭州奇异宇宙人工智能有限公司. Original upstream copyright notices, including OpenCode’s 2025 MIT notice, are retained separately.

License

Lyapunov Modified MIT, with retained upstream notices. Commercial products/services using Lyapunov or its derivatives must display prominent Lyapunov attribution at 10,000 monthly active users OR RMB 1,000,000 monthly revenue. The attribution clause does not automatically charge a license fee. See the actual license for its full terms.

中文版由杭州奇异宇宙人工智能有限公司出品。欢迎共同讨论、反馈问题和提交修正代码；当前为 Linux x64 Alpha，未来功能以路线图为准。

0.1.0α（Alpha2）

Lyapunov 0.1.0α (Alpha2) — Linux x64 Alpha

“To see a World in a Grain of Sand” — William Blake, Auguries of Innocence

Lyapunov is an LLM-first, 3D-native robotics and agent desktop framework, produced by Vorynel Co., Ltd.. It brings natural-language tools, local files, scene editing, robot preparation, controlled physics and camera observations into one workspace.

English README · 简体中文 · 中文安装与产品手册 · Download and system requirements

Install

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

Public release name: 0.1.0α. Technical version and release ID: 0.1.0-alpha.2; Git tag: v0.1.0-alpha.2. The default installer selects the version from the published manifest. The Linux x64 application includes Node.js, Electron, pinned DSH and product plugins; the default MuJoCo companion uses a private, version-specific Python environment. The installer verifies published archives, relocates the companion at its final path, checks dependencies and native physics, then switches current. Failed installation retains the previous version and user data. When needed, the installer requests normal operating-system authorization, prepares the missing desktop libraries or sandbox setup, and continues installation automatically; no additional manual setup command is required.

After installation reports readiness, start Lyapunov yourself from its desktop entry or a terminal:

lyapunov

If the user launcher directory is not on PATH, use "$HOME/.local/bin/lyapunov". The installer does not automatically open the application.

• Application archive: 551,722,973 bytes; SHA-256 410edf9f7f54657c38b7a6646419a5af7c13c0b2e06fbff0f8728fcfa8d1fc08.
• MuJoCo companion: 172,664,865 bytes; SHA-256 b8e090b9845a646ea7457f26f726757e1d40784603b55fe7ff2058cb19e033cb.
• Release manifest.

The versioned manifest records the actual source commit, archive sizes and SHA-256 values. The installer retains partial downloads and resumes them, then checks the actual bytes against that manifest. Previously published download URLs and objects remain unchanged.

What changed

Alpha 2 addresses the installation and workbench issues below:

• Graphical-library and sandbox preparation follows the actual desktop checks; the installer requests normal system authorization when needed and continues automatically after preparation.
• Existing desktop-entry information is kept for recovery instead of losing the previous entry during installation or repair.
• File-location displays use absolute paths and reuse the native file-browsing workflow.
• The mistakenly attached execution-graph button is removed from the default workbench UI.

The established robot, camera and local-workspace behavior is retained:

• Robot and world preparation use paused initialization followed by explicit, bounded execution. A source address or loaded SDK is distinct from verified control readiness.
• Saved camera views have a visible return to the main view. Camera binding, lens parameters, robot link frames and observations share native scene state.
• Static environment collision derivation preserves interior spaces with surface boxes; dynamic and explicit convex-decomposition paths retain their own behavior.
• Guest mode opens local features without product-account, model-relay or billing access. Agent use requires an explicitly configured personal provider.
• Robot catalog responses provide official source roots and metadata. The client uses browser, WebFetch and local tools to retrieve and check actual files; server metadata does not assert robot behavior.
• Continuous manual action feedback is consolidated, and scene/world selection, TCP, anchor and joint-control preparation remain visible in the workbench.

Documentation and examples

The English and Chinese READMEs cover environment and object construction methods, 19 input routes, robot import/download/compatibility, camera mounting and calibration, annotation editing, persistence and export. Four tutorials cover a ground/box scene, Panda joint/TCP controls, a mounted camera with annotations, and indoor collision alignment.

Native macOS, mobile/WeChat control, Genesis, VR/XR and companion DSH upgrades are listed in the roadmap. Windows uses the WSL2 + WSLg Linux-package route; customer WSLg acceptance remains pending, and this release does not claim complete Ubuntu 26 coverage. Native tools and APIs are recommended for routine workflows; CUA is not the default task interface.

Validation and Alpha scope

Source: 8a9c227902ca05c7d5af858f46cf6709015d11fd. The archive was built from this commit; the release tag additionally includes db5a7048452f086c126857c66fa29235d483b4ba, a one-line test-contract correction with no change to shipped runtime code. The standard package includes the updated English and Chinese READMEs; their actual archive bytes match this source. Use the archive identities recorded by the Alpha 2 manifest for this package. The document describes the fixes and retained scope; package, CI and runtime acceptance belong to their actual recorded Alpha 2 results.

This Alpha does not promise every GPU/distribution combination, full B1 navigation, arbitrary multi-robot collaboration, or G1 recovery from a fallen pose. Gait and get-up policies are separate capabilities. OS sandbox requirements are still checked by the installer; a successful desktop round is not universal platform acceptance.

Build together

Please use Discussions for ideas, Issues for reproducible problems, and pull requests for fixes. See CONTRIBUTING.

Thanks to DSH/Cordis, MuJoCo, NVIDIA Isaac Sim/IsaacLab, Three.js, Electron, Node.js, React, Blender, robot manufacturers, official model libraries and contributors. Their work and licenses are retained; Lyapunov contributes the integrated product workflow.

Authors and copyright

Mingjun Cheng; Zongjian Ding; Yudong Gao; Yi Yang; Lidong Chen; Jiale Liu; Xinling Yu.

© 2026 Vorynel Co., Ltd. The Chinese producer remains 杭州奇异宇宙人工智能有限公司. Original upstream copyright notices, including OpenCode’s 2025 MIT notice, are retained separately.

License

Lyapunov Modified MIT, with retained upstream notices. Commercial products/services using Lyapunov or its derivatives must display prominent Lyapunov attribution at 10,000 monthly active users OR RMB 1,000,000 monthly revenue. The attribution clause does not automatically charge a license fee. See the actual license for its full terms.

中文版由杭州奇异宇宙人工智能有限公司出品。欢迎共同讨论、反馈问题和提交修正代码；当前为 Linux x64 Alpha，未来功能以路线图为准。

Lyapunov 0.1.0α — Linux x64

Lyapunov 0.1.0α

An LLM-first, 3D-native robotics and agent desktop framework, produced by 杭州奇异宇宙人工智能有限公司.

English README · 简体中文 · 中文产品与安装手册

Install

curl -fsSL https://vorynel.com/lyapunov/install.sh | sh

The installer downloads and verifies the application and its private MuJoCo/Python companion, creates the launcher and desktop entry, and prints launch instructions. It does not automatically open the application. Node.js, Electron and pinned DSH are included. Follow the exact OS sandbox setup instructions when reported.

Downloads

Technical version and release ID: 0.1.0-alpha.1. Source: 1f5c320d88a38072acb0c40af7938b159a55c2b0.

• Linux x64 application: 551,017,689 bytes; SHA-256 0aabcf321b282fdef52c88b8ce068d135c991667f7efd1c7d50929f683f3dd13.
• MuJoCo companion: 172,664,865 bytes; SHA-256 b8e090b9845a646ea7457f26f726757e1d40784603b55fe7ff2058cb19e033cb.

Large archives are served through OSS and the existing download CDN. The website serves the installer and release metadata. Both CDN archives passed full byte/hash checks and Range 206 checks. Observed download rates at one test location were 36.2 MB/s and 34.7 MB/s; client performance varies by network and location.

This release

This release unifies the actual package, desktop, archive filenames, JSON/TSV manifests and current documentation under the Alpha version. It retains the A08 fixes for paused robot preparation, visible camera return, static environment collisions, consolidated manual-action feedback, local Guest mode and client-owned source retrieval. See the full release notes.

Actual CI passed at the exact source commit: 333/333 registered files passed, with the existing exclusions retained. The rebuilt package version and source identity were independently read back. The default customer installation and VM user-interface round remain under manual acceptance; previous runtime checks do not substitute for that round.

Scope and community

Linux x64 Alpha. WSL2/WSLg real-machine acceptance, native macOS, mobile/WeChat control, Genesis, VR/XR and companion DSH upgrades remain on the roadmap. Gait and get-up policies are separate capabilities; this release does not claim arbitrary multi-robot collaboration or full indoor navigation.

Please share ideas in Discussions, report reproducible problems in Issues, and contribute fixes through pull requests.

Thanks to DSH/Cordis, MuJoCo, NVIDIA Isaac Sim/IsaacLab, Three.js, Electron, Node.js, React, Blender, robot manufacturers, official model libraries and contributors. Their licenses and notices are retained.

Lyapunov Modified MIT: commercial products/services using Lyapunov or its derivatives must display prominent Lyapunov attribution at 10,000 monthly active users OR RMB 1,000,000 monthly revenue. See LICENSE for the complete terms.

中文版由杭州奇异宇宙人工智能有限公司出品。欢迎讨论、反馈与提交修正代码；本版本为0.1.0α，技术标识0.1.0-alpha.1。