ABOUT ZERAIX AI

为智能原生时代,构筑端侧基础。

原生纪(Zeraix AI)专注于端侧智能软件与基础设施,团队成员来自清华大学、浙江大学等院校。我们以自研推理引擎与开源智能体为核心,连接模型、设备与真实业务,致力于让 AI 成为持续创造价值的生产力。

  • 端侧基础设施
  • 开源创新
  • 产业 AI 化
浅色立体插画:蓝青与淡紫光带连接开放的智能网络

01 / OUR BELIEFS

让智能,在更多地方生长。

我们相信,AI 的价值不仅在于模型有多强,更在于有多少人、多少设备、多少业务能够真正使用它。

01

智能普惠

让智能更接近使用它的人。通过端侧运行与持续优化,探索更易获得、更高效、更可控的 AI 使用方式。

02

开放创新

以自研技术积累底层能力,以开放源码连接开发者。让实现可以被理解、验证,也可以被更多人共同改进。

03

产业共进

从业务深处寻找 AI 的价值,把模型能力转化为可运行的产品与工作流程,与企业共同探索新的生产方式。

02 / OUR VISION

从运行模型,到支撑智能。

我们的长期愿景,是为处理复杂任务的端侧智能设备构建通用执行底座。让有限的计算与存储资源,支撑更持续、更协同的智能工作。

  1. 深耕模型执行

    围绕推理效率、硬件适配与状态复用,持续打磨 Imparo 的执行能力。

  2. 连接任务与工具

    以 Zeraix Agent 为工作入口,探索文件、工具与多步骤任务的协同。

  3. 走向端侧协同

    面向更复杂的设备与任务,持续探索资源调度、状态管理与多任务执行。

以今天的产品积累,走向长期愿景。通用执行底座是持续研发方向,而非已经完成的产品承诺。

03 / OPEN SOURCE

让创新,发生在开放之中。

两个独立迭代的开源项目,分别深耕智能体工作空间与模型执行。欢迎使用、验证,也欢迎共同改进。

Zeraix Agent

本地优先的开源智能体工作空间,连接模型、文件与工具,让 AI 参与真实的多步骤工作。

了解 Zeraix Agent

Imparo

自研开源推理引擎,围绕硬件、负载与状态复用持续优化,让模型更好地利用设备资源。

了解 Imparo

04 / BUILD TOGETHER

与同行者,共建智能原生生态。

面向企业、硬件伙伴与开发者开放合作,把不同的行业经验与技术能力,汇聚成可以落地的创新。

01

企业共创

深入梳理业务,从零到一共建专属 AI 应用与工作流程。

02

硬件协同

围绕设备形态、算力约束与实际任务,探索端侧智能产品。

03

开源共建

参与功能开发、模型适配与问题反馈,让工具在共同使用中完善。

04

研究交流

围绕推理、缓存与调度等方向,交流方法、实验与可复现结果。

查看公开能力、适配边界与资料来源

两个项目独立迭代:Zeraix 当前通用推理采用 llama.cpp 系运行时,Imparo 尚非桌面端默认内置引擎。具体功能以对应版本为准。

Zeraix Agent:面向文件、工具与多步任务的开源智能体工作空间。

Imparo:面向硬件与负载自适应执行的开源推理引擎,持续探索原生执行、缓存复用与状态管理。

从技术到产业:深入企业业务流程,从零到一设计并搭建 AI 应用;面向硬件团队开展模型适配与推理优化,让技术与实际需求共同演进。

产品能力以对应仓库与发布版本为准;长期愿景与研究方向不代表现阶段已完成的产品功能。

与我们一起,拓展智能的可能。

无论你在构建新的业务、新的设备,还是新的技术,欢迎与原生纪交流。

联系我们

ABOUT ZERAIX AI

Foundations foran AI-native world.

Zeraix AI builds software and infrastructure for intelligence at the edge. Our team includes members from Tsinghua University, Zhejiang University and other institutions. With our own inference engine and an open-source agent workspace, we connect models, devices and business workflows to turn AI into lasting value.

  • Edge infrastructure
  • Open-source innovation
  • AI for industry
Light 3D illustration of an open network connected by blue, cyan and lilac ribbons

01 / OUR BELIEFS

Intelligence belongs wherever life and work happen.

The value of AI lies not only in what a model can do, but in how many people, devices and businesses can put it to use.

01

Make intelligence accessible

Bring intelligence closer to the people who use it. Explore accessible, efficient and controllable AI through on-device execution and continued optimization.

02

Build in the open

Develop our own core technology and share it through open source, so developers can understand, validate and improve the work.

03

Grow with industry

Start inside the business. Turn model capabilities into working products and workflows, and explore new ways of working with the people who use them.

02 / OUR VISION

From running models to supporting intelligence.

Our long-term ambition is to build a general-purpose execution foundation for edge devices handling complex tasks—making sustained, coordinated AI work possible within limited compute and memory.

  1. Advance model execution

    Develop Imparo around inference efficiency, hardware adaptation and reusable execution state.

  2. Connect tasks and tools

    Explore coordinated work across files, tools and multi-step tasks through the Zeraix Agent workspace.

  3. Work toward coordinated edge AI

    Research resource scheduling, state management and multi-task execution for more complex devices and workloads.

Today’s products are steps toward this ambition. A general-purpose execution foundation remains a research direction, not a completed product claim.

03 / OPEN SOURCE

Build openly. Move forward together.

Two independently evolving projects, focused on the agent workspace and model execution. Use them, test them and help improve them.

Zeraix Agent

A local-first, open-source workspace connecting models, files and tools for real multi-step work.

Discover Zeraix Agent

Imparo

Our open-source inference engine, developed around hardware, workloads and reusable state to make better use of device resources.

Discover Imparo

04 / BUILD TOGETHER

Build an AI-native ecosystem, together.

We welcome enterprises, device teams and developers to turn complementary expertise into practical innovation.

01

Enterprise co-development

Map the business and build custom AI applications and workflows from the ground up.

02

Device collaboration

Explore on-device AI products around hardware constraints, form factors and real tasks.

03

Open-source contributions

Contribute features, model support and feedback to improve tools through shared use.

04

Research exchange

Exchange methods, experiments and reproducible findings on inference, caching and scheduling.

Published capabilities, qualification boundaries and sources

The projects evolve separately. Zeraix currently uses llama.cpp-based runtimes for general inference; Imparo is not yet its bundled default engine. Features depend on the selected version.

Zeraix Agent:An open-source workspace for files, tools and multi-step agent tasks.

Imparo:An open-source inference engine exploring hardware- and workload-adaptive execution, context reuse and state management.

From technology to industry:Understand business workflows and build custom AI applications from the ground up. Work with device teams on model adaptation and inference performance, guided by actual needs.

Product capabilities follow the relevant repositories and releases. Long-term ambitions and research directions are not claims of completed features.

Help shape what intelligence makes possible.

Whether you are building a business, a device or a new technology, we would like to hear from you.

Contact us