A configurable machine platform
The design combines a Linux substrate, Rust-owned runtime, hardware interfaces and permissioned applications, with full-OS and existing-Linux integration profiles.
商业摘要已提供本语言版本。产品详细流程和技术证据以英语维护。
物理 AI 与机器人
面向 Physical AI 与工业 AI 的操作系统平台。
每一种新机器都可能需要重新构建模型集成、权限、设备接口、更新和机群运营。哪些部分应该成为可复用的平台?
面向机器制造商、工业 OEM 和机器人平台负责人
开发中 · 平台设计
Permission is not proof of execution. The design keeps AI proposals, machine authority and independent physical protection separate.
An application or model proposes work through the platform API. Identify the machine, intended effect, owner and operating constraints.
Permission expires. An expired lease cannot authorise a new action. Re-evaluate the request and current state; expiry is not completion.
Local veto. Preserve the controller’s refusal. The platform cannot overrule it or report execution.
Read back. Verify observed state and retain the journal. Reconcile ambiguity before considering another request; permission alone never proves success.
Designed action path · authority, device control and observed state
源码审查日期
The decision behind the system
Auralink is being designed around a Rust-owned core: configurable machine images, AI applications, deterministic on-device authority and vendor interfaces. Local open-weight inference and optional fleet services belong to the same platform design.
The next Auralink design extends beyond coordination into a machine operating system and application platform. Its Rust-owned core, domain packs and optional hub are design work; the predecessor's implementation records remain separate.
The design combines a Linux substrate, Rust-owned runtime, hardware interfaces and permissioned applications, with full-OS and existing-Linux integration profiles.
AI proposes work. A deterministic on-device service is designed to check identity, policy and current state; independent protective functions retain their own authority.
Recording, replay, model evaluation, signed updates and compatibility checks are planned alongside domain packs and an optional self-hosted fleet hub.
An application or model proposes work through the platform API. Identify the machine, intended effect, owner and operating constraints.
The on-device authority service is designed to check identity, permissions, policy and state freshness before granting bounded authority.
The device interface must enforce the permitted scope. Local controllers can refuse; the safety island and hardware protective functions have separate responsibilities.
Observed state and journals must distinguish completion, refusal and ambiguity. Recording and replay support investigation; a permission never proves physical success.
Inspect the source, test conditions and operating limits in the technical record.
Platform architecture · design baseline
Auralink is being designed as an operating-system platform for Physical and Industrial AI. The next version combines a configurable Linux foundation with a Rust-owned core, a permissioned application model and vendor interfaces. Implementation has not started.
Permissioned applications and agents, local open-weight inference and policy-selected hosted models. An offline profile is part of the design.
A deterministic service evaluates proposals against identity, policy, permissions and current state. Models and fleet services do not acquire machine authority.
The Rust-owned runtime and vendor interfaces connect applications to machine capabilities. The safety-island design and hardware protective functions have independent responsibilities.
Kconfig selections, Buildroot images and a real-time Linux foundation define what belongs on each machine. Adopted components remain outside the Rust-owned core.
The design includes recording and replay, model evaluation and shadow operation, signed updates, compatibility checks and rollback paths. Each needs implementation and verification before release.
The target profile controls the machine image, boot, kernel, updates and runtime. Its guarantees depend on the selected hardware and completed qualification.
An integration path for machines that retain their existing OS. It has a narrower control boundary and cannot inherit the full-OS guarantees.
A self-hostable hub is planned for enrolment, identity, tenant boundaries, audit, rollout, data and operator interfaces. Domain back-office modules sit outside the core. Machines are designed to keep local functions available without the hub; it holds no machine-authority keys.
Planned domain packs
Four planned domain packs separate protocol and application work from the reusable platform. Each pack needs its own implementation, interoperability and hardware evidence.
A planned charging pack for charger interfaces, energy constraints and operator workflows. Optional hub modules cover central-system and back-office responsibilities.
Robot and fleet interfaces, bounded machine actions and shared-resource coordination. Motion control and physical protection retain explicit owners.
A planned pack for infrastructure interfaces and constrained coordination across public assets. Each installation needs its own operational and safety assessment.
Industrial interfaces and permissioned applications over a common machine platform. Existing PLC, SCADA and protective-system responsibilities remain explicit.
Pack selection does not establish device support. The target machine, vendor interface and operating environment determine the qualification work.
业务 ↔ 技术桥梁 · Physical AI platform design
Connects machine-platform product decisions to a Rust-owned operating-system design and its implementation and qualification plan.
这是对项目组合与源代码的解读,并非客户、部署、认证或成果声明。
Every new physical-agent vertical can become another bespoke platform, multiplying integration, operations, support, and assurance cost.
Choose what belongs in the machine platform, domain pack, vendor interface and optional fleet hub, with explicit authority and integration-profile boundaries.
A design baseline for configurable images, a Rust-owned core, on-device authority, local inference, application permissions, domain packs, signed updates and fleet services.
Runtime constraints and validation gaps reopen platform scope, supported-vertical claims, deployment obligations, release decisions, and product economics.
本页解读的是独立项目仓库中的源代码。源代码存在,并不表示该项目作为服务运行在 hyperion-consulting.io 内。
访客及客户数据的 AI 处理限于通过受治理的欧盟端点使用 Mistral。所有者授权的 Helio、搜索及 Product Council 流程已启用。独立的公开来源编辑流程使用 Anthropic,并以受治理的 Mistral 作为备用;该流程不接收网站咨询、客户文档或访客对话。远程 AI 不可用时,确定性工具仍可使用。
源码证据
The next Auralink is a proposed operating-system and application platform for Physical and Industrial AI machines. The reviewed design covers a Rust-owned core, configurable Linux images, deterministic on-device authority, model integration, domain packs and an optional fleet hub.
详细技术快照以英文维护。
代码仓库证据不等同于客户、部署、认证或成果证据。
Start with one machine, one application and named owners. Establish the implementation and acceptance evidence before considering any physical trial.
Select the business outcome, target machine and full-OS or existing-Linux integration profile. Name the product, engineering and operating owners.
Review conflicts, policy, data, hazards, economics and the evidence needed to make a decision.
Plan approved replay data, simulation and read-only observation to evaluate the proposed authority and model boundaries once implemented.
After implementation, test expiry, local veto, ambiguous outcomes, readback and recovery against an authorised virtual interface.
Consider hardware-in-the-loop or a bounded pilot only after separate hardware, security, safety and operator gates are met.
Implementation and qualification
The next platform starts from an architecture and planning baseline. Progress depends on implementation and demonstrated behaviour, with separate software, interoperability and physical qualification gates.
Build the configurable image, Rust-owned runtime, on-device authority and reference interfaces. Verify denial, recovery and update boundaries.
Implement recording and replay, model routing and permissioned applications. Compare model behaviour against task-specific acceptance criteria.
Exercise domain packs and vendor interfaces against independently implemented peers, including failures and degraded operation.
Use target hardware, measured timing, fault injection and named operator acceptance before considering a bounded field trial.
Model output must not grant itself access to an actuator. On-device checks, local controllers and independent protective functions have different roles to verify.
Earlier Auralink software and simulation results inform the replacement design. They do not qualify the next implementation, its hardware interfaces or any future release.
These technical documents, source-access paths and authorized research results concern the predecessor. They inform the next design; they do not establish that the new platform is implemented or qualified.
工作原理
Applications propose work through vertical interfaces and shared capabilities to the horizontal coordination runtime. Device, edge and cloud describe where the work runs. The technical reference explains the coordination specification and its EV-charging research lineage.
深入了解
深入了解 Auralink。访问需签署 NDA。
这些文档包含专有架构细节、基准测试和实现细节。一份简短的 NDA 可保护双方权益。
为何协调层是分布式物理 AI 缺失的那个原语,以及它的设计旨在跨 EV / 机器人 / 智慧城市 / IIoT 领域实现什么。
ACS v0.1——六种拟议的规范性资源类型、四个安全等级、平滑递减式抢占、差分隐私下的联邦学习。附 TLA+ 形式化模型。
横跨云端与边缘的完整系统架构、用 Rust 编写的充电桩 OS、用 Go 编写的智能体捆绑包、部署拓扑、RAG 主干,以及作为已记录设计范围的超大型项目实施手册。
源代码
Request access to source repositories and any configured model artefacts under the applicable agreements.
请先签署 NDA,再申请代码访问权限。
与 Hyperion 合作
有明确范围的架构审查:目标机器、集成配置、权限边界,以及待建立的实现与适用性验证证据。
请简要说明您的情况、限制条件和待作决策。开始工作前,我们会商定范围、责任和报价。