diff --git a/README.md b/README.md index 395dcde..51d6fd2 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,16 @@ + +> [!NOTE] +> 本文档由 WeHub 基于上游 README 翻译整理，属于社区翻译，非官方中文文档。 +> [English](./README.en.md) · [原始项目](https://github.com/Mininglamp-AI/Mano-P) · [上游 README](https://github.com/Mininglamp-AI/Mano-P/blob/HEAD/README.md) +> 原作者、版权与许可证归属以原始项目及本仓库 LICENSE 文件为准。 +

Mano-P Logo Mano-P 2.0

-

GUI-Aware Agent Model for Edge Devices

-

Private AI

+

面向端侧设备的GUI感知智能体模型

+

私有 AI

@@ -18,9 +24,9 @@ [![ModelScope CN](https://img.shields.io/badge/🪄-ModelScope%20CN-purple)](https://www.modelscope.cn/models/Mininglamp2718/Mano-CUA-2.0-4B) [![ModelScope AI](https://img.shields.io/badge/🪄-ModelScope%20AI-purple)](https://www.modelscope.ai/models/Mininglamp2718/Mano-CUA-2.0-4B) -中文 | English +中文 | English -**[📖 Overview](#-overview) | [🎯 Key Highlights](#-key-highlights) | [🎬 Use Cases](#-use-case-demonstrations) | [📊 Benchmark](#-benchmark-performance) | [🔧 Skills](#-skills) | [🤖 Models](#-models) | [⚡ Inference SDK](#-inference-sdk) | [⚗️ Approach](#-approach) | [🌟 Technical Advantages](#-technical-advantages) | [🔗 Applications](#-applications) | [📄 Citation](#-technical-papers--citation) | [❓ FAQ](#-faq)** +**[📖 项目概述](#-项目概述) | [🎯 核心亮点](#-核心亮点) | [🎬 应用场景](#-应用场景展示) | [📊 基准测试](#-基准测试性能) | [🔧 Skills](#-skills) | [🤖 模型](#-模型) | [⚡ 推理加速 SDK](#-推理加速-sdk) | [⚗️ 方法](#-方法) | [🌟 技术优势](#-技术优势) | [🔗 应用](#-应用) | [📄 论文引用](#-技术论文与引用) | [❓ FAQ](#-常见问题)** @@ -34,94 +40,93 @@ --- -## 📖 Overview +## 📖 项目概述 -**Mano-P**: "Mano" means "hand" in Spanish, and "P" stands for Private. We believe that both individuals and organizations can create their own Private AI, and a bright future of human-machine collaboration is on the horizon. +**Mano-P**，Mano 是西班牙语里"手"的意思，P 代表 Private。我们相信，个体和组织都能够创造属于自己的私有 AI，人机协同的美好世界即将到来。 -![opensource_architecture.png](pics/opensource_architecture_en.png) +![opensource_architecture.png](pics/opensource_architecture_zh.png) -**Mano-P** is a GUI-VLA agent project designed specifically for edge devices. It serves both as an open-source project and a hardware product solution. -As an open-source project, Mano-P is being released in a phased, progressive manner, targeting three distinct groups of developers. In the first phase, we will open-source the Mano-CUA Skills. This phase is aimed at Agent enthusiasts—such as users of OpenClaw or Claude Code—enabling them to leverage the capabilities of Mano-CUA Skills to construct more intelligent CUA task workflows and overcome the bottlenecks associated with human intervention. In the second phase, we will open-source the local-side models and SDK components of Mano-CUA. This phase targets developers with high security requirements, allowing them to directly utilize GUI-VLA models capable of running inference locally on a Mac mini to build their own custom Skills, Tools, and more; **crucially, all your CUA operations will be executed entirely on your local Mac mini and will not be uploaded to external servers.** In the third phase, we will open-source the training methodologies and the pruning and quantization techniques used for the Mano-P models. This phase is designed for developers with specific model training needs, empowering them to apply our training methods to create their own on-device GUI-VLA models tailored to their unique requirements. +**Mano-P** 是一个专为边缘设备设计的 GUI-VLA 代理项目。它既是一个开源项目，也是一个硬件产品解决方案。作为开源项目，Mano-P 将面向三类不同的开发者群体，分阶段逐步发布。在第一阶段，我们将开源 Mano-CUA 技能。此阶段的目标用户是Agent爱好者，例如 OpenClaw 或 Claude Code 的用户，使他们能够利用 Mano-CUA 技能的功能构建更智能的 CUA 任务工作流程，并克服人工干预带来的瓶颈。在第二阶段，我们将开源 Mano-CUA 的本地模型和 SDK 组件。此阶段的目标用户是具有高安全性要求的开发者，使他们能够直接使用可在 Mac mini 本地运行推理的 GUI-VLA 模型来构建自定义技能、工具等。**重要的是，所有 CUA 操作都将在本地 Mac mini 上执行，而不会上传到外部服务器**。在第三阶段，我们将开源 Mano-P 模型所使用的训练方法、剪枝和量化技术。此阶段旨在满足开发者特定的模型训练需求，使他们能够应用我们的训练方法，创建符合自身独特需求的本地 GUI-VLA 模型。 -Regarding our GUI-VLA models—which are capable of running inference directly on Mac mini and MacBook devices—we currently support two deployment methods: First, direct deployment on Mac mini or MacBook models equipped with an M4 chip and 32GB or more of RAM; and second, deployment utilizing a compute stick connected via a USB 4.0 port or higher. We will be releasing detailed instructions for both deployment methods in the near future, and we plan to expand our support to include additional deployment options in the future. +关于我们的 GUI-VLA 模型（可在 Mac mini 和 MacBook 设备上直接运行推理），我们目前支持两种部署方式：第一种是直接部署在配备 M4 芯片和 32GB 或以上内存的 Mac mini 或 MacBook 机型上；第二种是使用通过 USB 4.0 或更高版本端口连接的算力棒进行部署。我们将在近期发布这两种部署方式的详细说明，并计划在未来扩展支持范围，纳入更多部署选项。 -### Main Capabilities +### 主要能力 -- **Complex GUI Automation**: Autonomously complete complex interface operations containing hundreds of interactive elements -- **Cross-System Data Integration**: Extract and integrate multi-source data through pure visual interaction without API interfaces -- **Long-Task Planning Execution**: Support enterprise-level business process automation of dozens to hundreds of steps -- **Intelligent Report Generation**: Automatically generate structured documents such as data analysis reports and work summaries -- **Edge-Native Inference**: Efficient on-device execution on Apple Silicon via INT8 activation quantization ([Cider](#-inference-sdk)) -- **Autonomous Application Construction**: Drives end-to-end software construction pipelines through visual GUI operation ([Mano-AFK](#-applications)) +- **复杂 GUI 自动化操作**：自主完成包含数百个交互元素的复杂界面操作 +- **跨系统数据整合**：无需 API 接口，通过纯视觉交互提取和整合多源数据 +- **长任务规划执行**：支持数十步至上百步的企业级业务流程自动化 +- **智能报告生成**：自动生成数据分析报告、工作总结等结构化文档 +- **端侧原生推理**：基于 INT8 激活量化在 Apple Silicon 上高效本地执行（[Cider](#-推理加速-sdk)） +- **自主应用构建**：以视觉 GUI 操作驱动端到端软件构建流水线（[Mano-AFK](#-应用)） -### Technical Background +### 技术背景 -Mano-P builds upon the complete technical framework of the Mano project (see [Mano Technical Report](https://arxiv.org/abs/2509.17336)), employing the Mano-Action bidirectional self-reinforcement learning method, three-stage progressive training (SFT → Offline Reinforcement Learning → Online Reinforcement Learning), "think-act-verify" loop reasoning mechanism, and a closed-loop data circulation system to achieve high-precision GUI understanding and operation capabilities. The edge version is optimized through mixed-precision quantization, visual token pruning, and edge inference adaptation, enabling large-scale parameter models to run efficiently on edge devices like Mac mini/MacBook/computing sticks. +Mano-P 基于完整的 Mano 项目技术体系（详见 [Mano Technical Report](https://arxiv.org/abs/2509.17336)），采用 Mano-Action 双向自增强学习方法，通过三阶段渐进式训练（SFT → 离线强化学习 → 在线强化学习）和"思考-行动-验证"循环推理机制，配合闭环数据循环系统，实现了高精度的 GUI 理解和操作能力。端侧版本通过混合精度量化、视觉 Token 剪枝和边缘推理自适应等优化，使大参数量模型能够在 Mac mini/MacBook/算力棒等端侧设备上高效运行。 -## 🎯 Key Highlights +## 🎯 核心亮点 -- **#1 on OSWorld Benchmark**: Mano-CUA 1.1 achieves **58.2% success rate on OSWorld**, ranking first among all specialized GUI agent models, outperforming the second-place opencua-72b (45.0%) by 13.2 percentage points -- **Leading on WebRetriever Protocol I**: Mano-CUA 1.1 scores **41.7 NavEval**, surpassing Gemini 2.5 Pro Computer Use (40.9) and Claude 4.5 Computer Use (31.3) -- **Fully Local Execution**: Runs inference locally on **Apple M4 chip with 32GB RAM** (Mac mini or MacBook). No cloud API calls required. All screenshots and task data stay on-device -- **High-Performance Inference**: Mano-CUA-4B achieves **~80 tokens/s decode** on Apple M5 Pro; with Cider's W8A8 activation quantization, prefill speeds up by **~12.7%** over the W8A16 baseline -- **Autonomous Long-Task Execution**: Supports **complex business processes** with end-to-end automation without internet connectivity -- **Edge-Native INT8 Acceleration**: Companion [Cider](#-inference-sdk) SDK adds the W8A8 / W4A8 activation-quantization primitives MLX lacks natively, delivering **1.4x–2.2x prefill speedup** over MLX W4A16 on Apple M5 Pro — works with any MLX model, not just Mano-P -- **Autonomous Software Construction**: [Mano-AFK](#-applications) drives a full PRD → code → deploy → test → fix loop using Mano-P as its local vision model for real-browser E2E testing — from a single natural-language prompt to a deployed, tested application, no human in the loop +- **OSWorld 基准测试第一**：Mano-CUA 1.1 在 OSWorld 上取得 **58.2% 成功率**，在所有专用 GUI 智能体模型中排名第一，领先第二名 opencua-72b (45.0%) 达 13.2 个百分点 +- **WebRetriever Protocol I 领先**：Mano-CUA 1.1 取得 **41.7 NavEval 分数**，超越 Gemini 2.5 Pro Computer Use (40.9) 和 Claude 4.5 Computer Use (31.3) +- **完全本地运行**：在**苹果 M4 芯片 + 32GB 内存**的 Mac mini/MacBook 上本地推理，无需云端 API，所有截图和任务数据不出设备 +- **高性能推理**：Mano-CUA-4B 在 Apple M5 Pro 上实现 **~80 tokens/s 解码**；配合 Cider 的 W8A8 激活量化，prefill 相对 W8A16 baseline 加速约 **12.7%** +- **长任务自主执行**：支持**复杂业务流程**的端到端自动化，无需联网 +- **端侧 INT8 加速**：配套 [Cider](#-推理加速-sdk) SDK 补齐了 MLX 原生缺失的 W8A8 / W4A8 激活量化原语，在 Apple M5 Pro 上相对 MLX 原生 W4A16 实现 **1.4x–2.2x 的 prefill 加速**——兼容任意 MLX 模型，并非 Mano-P 专属 +- **自主软件构建**：[Mano-AFK](#-应用) 以 Mano-P 为本地视觉模型驱动真实浏览器 E2E 测试，打通 PRD → 代码 → 部署 → 测试 → 修复的完整闭环，从一句自然语言描述到可运行、已测试的应用，全程零人工介入 --- -## 🎬 Use Case Demonstrations +## 🎬 应用场景展示 -### Scenario 1: Mano-AFK Fully automated application construction +### 场景 1: Mano-AFK 全自动化应用构建 -https://github.com/user-attachments/assets/8512ab65-f836-4779-979a-4c636fe61fd2 +https://github.com/user-attachments/assets/7637957d-aa5e-48c1-b823-56ff392181ab -We demonstrated the fully automated application construction process of [Mano-AFK](#-applications). After receiving natural language requirements, the system sequentially completes requirement clarification, technical architecture design, code generation, local deployment, and multi-level testing (API interface testing, LLM based page visual inspection, and end-to-end GUI automation testing driven by VLA model). When the test fails, the system automatically locates the root cause of the problem, fixes the code, and deploys verification again, iterating until all test cases pass. The entire process does not require manual intervention, and ultimately delivers a runnable application with complete requirement documents and build reports. +我们演示了 [Mano-AFK](#-应用) 全自动化应用构建流程。系统接收自然语言需求后，依次完成需求澄清、技术架构设计、代码生成、本地部署及多层级测试（API接口测试、基于LLM的页面视觉检测、以及通过VLA模型驱动的端到端GUI自动化测试）。测试未通过时，系统自动定位问题根因、修复代码并重新部署验证，循环迭代直至所有测试用例通过。全流程无需人工干预，最终交付可运行的应用及完整的需求文档与构建报告。 -[![Watch on YouTube](https://img.shields.io/badge/Watch%20on-YouTube-red?logo=youtube)](https://youtu.be/T2QeXOOvRBQ?si=-I1HDmmtWNeKmg5Q) +[![Watch on 微信视频号](https://img.shields.io/badge/Watch%20on-微信视频号-07C160?logo=wechat&logoColor=white)](https://weixin.qq.com/sph/A7eBqGJkH0) -### Scenario 2: Commercial video intelligent system +### 场景 2: 商业视频智能系统 -https://github.com/user-attachments/assets/04730188-e664-4f92-8ba7-023269880718 +https://github.com/user-attachments/assets/64c7dca1-973f-4c36-b30e-1f4e0e9e8a03 -We fully demonstrated the actual workflow of a commercial video intelligent system. Starting from the user's command, the system automatically completes the entire process of video generation, uploading, analysis, editing, and secondary evaluation. During the process, the system can autonomously operate web pages and editing software, complete fine operations such as file processing and subtitle modification, and generate analysis reports containing subjective evaluations and objective indicators. By comparing the differences between the initial and refined versions, visually present the overall capabilities and application effects of the system. +我们完整演示了一套商业视频智能系统的实际工作流程。从用户下发指令开始，系统自动完成视频生成、上传、分析、剪辑再到二次评测的全过程。过程中，系统可自主操作网页与剪辑软件，完成文件处理、字幕修改等精细操作，并生成包含主观评价与客观指标的分析报告。通过对比初版与精剪版本的差异，直观呈现系统的整体能力与应用效果。 -[![Watch on YouTube](https://img.shields.io/badge/Watch%20on-YouTube-red?logo=youtube)](https://youtu.be/g4sXOTtNPbo?si=RmV5wLLlI1u4e7Nj) +[![Watch on 微信视频号](https://img.shields.io/badge/Watch%20on-微信视频号-07C160?logo=wechat&logoColor=white)](https://weixin.qq.com/sph/A0vd7EAezv) -### Scenario 3: Local model task execution +### 场景 3: 本地模型任务执行 -https://github.com/user-attachments/assets/992f4961-3028-45c9-a7c2-29a8e5bf93a9 +https://github.com/user-attachments/assets/cb3e65be-eaf5-44a5-9f38-1415c12a8a43 -Mano-P, The small-sized end side GUI-VLA model can run directly on your computer, supporting direct inference operation on Macmini/Macbook with M4 chip and above, as well as direct operation on plug and play computing power sticks. In the CUA scenario, break through the bottleneck of human participation in the Agent workflow. Mano-P, The first step in leading Private AI. +Mano-P，小尺寸端上GUI-VLA模型，直接运行在你的电脑上，支持M4芯片及以上的Macmini/Macbook直接推理运行，也支持在即插即用算力棒上直接运行。在CUA场景中，打通Agent工作流人类参与其中的瓶颈。Mano-P，引领私有 AI 第一步。 -[![Watch on YouTube](https://img.shields.io/badge/Watch%20on-YouTube-red?logo=youtube)](https://youtu.be/VyHhsO1HFpg) +[![Watch on 微信视频号](https://img.shields.io/badge/Watch%20on-微信视频号-07C160?logo=wechat&logoColor=white)](https://weixin.qq.com/sph/AkCLmX4NTA) -### Scenario 4: Daily Life and Entertainment Applications +### 场景 4: 生活娱乐场景应用 -https://github.com/user-attachments/assets/ff11fd5b-9ee7-4a74-b8e6-3ad3071d3af8 +https://github.com/user-attachments/assets/397a0552-9611-4d74-9f24-99544da272b6 -Mano-P excels not only in enterprise-level business automation but also integrates seamlessly into daily life. This video demonstrates the system's application in Mahjong gameplay: through pure visual understanding of the game interface, it autonomously completes tile recognition, analysis, and decision-making. This case validates Mano-P's general-purpose capabilities beyond work scenarios—from office automation to leisure entertainment, from structured data processing to unstructured game interactions, truly realizing the vision of "Private AI." One model, adapting to every aspect of life and work. +Mano-P不仅能胜任企业级业务自动化，更能融入日常生活。本视频展示系统在麻将游戏中的应用：通过纯视觉理解游戏界面，自主完成识牌、分析和决策。这一案例验证了Mano-P在非工作场景下的通用能力——从办公自动化到休闲娱乐，从结构化数据处理到非结构化游戏交互，真正实现"私有 AI"的愿景。一个模型，适配生活与工作的方方面面。 -[![Watch on YouTube](https://img.shields.io/badge/Watch%20on-YouTube-red?logo=youtube)](https://youtu.be/P4-wE3p7mB8) +[![Watch on 微信视频号](https://img.shields.io/badge/Watch%20on-微信视频号-07C160?logo=wechat&logoColor=white)](https://weixin.qq.com/sph/AOzBArrvd7) -### Scenario 5: Mano-AFK × Cider Locally-Accelerated End-to-End App Construction +### 场景 5: Mano-AFK × Cider 本地加速端到端应用构建 -https://github.com/user-attachments/assets/b987f367-5c47-4659-890b-3e8a6e741603 +https://github.com/user-attachments/assets/5215d4eb-4e6f-4e03-b31b-dc8037c3794d -This video demonstrates the combined capability of [Mano-AFK](#-applications) and the [Cider](#-inference-sdk) inference acceleration SDK. Starting from a single natural-language requirement, Mano-AFK autonomously performs requirement clarification, architecture design, code generation, and local deployment. In the E2E testing stage it drives a real browser through the local Mano-P vision model — accelerated by Cider — to run GUI automation tests; on failure, it localizes the defect, patches the code, and re-verifies until a runnable application is delivered. Cider supplies the INT8 activation-quantization primitives that give Mano-P a substantial prefill speedup on Apple Silicon, so the entire build–test–fix loop runs fully on-device, balancing autonomy, privacy, and performance. +本视频展示 [Mano-AFK](#-应用) 与 [Cider](#-推理加速-sdk) 推理加速 SDK 的联合能力。Mano-AFK 从一句自然语言需求出发，自主完成需求澄清、架构设计、代码生成、本地部署，并在 E2E 测试环节调用由 Cider 加速的本地 Mano-P 视觉模型驱动真实浏览器完成 GUI 自动化测试；测试失败时自动定位缺陷、修复代码并重新验证，直至交付可运行的应用。Cider 提供 INT8 激活量化原语，让 Mano-P 在 Apple Silicon 上获得显著的 prefill 加速，整个"构建—测试—修复"闭环完全在本地执行，兼顾自主性、隐私与性能。 -[![Watch on YouTube](https://img.shields.io/badge/Watch%20on-YouTube-red?logo=youtube)](https://youtu.be/PK-QYjiAkb8) +[![Watch on 抖音](https://img.shields.io/badge/Watch%20on-抖音-000000?logo=tiktok&logoColor=white)](https://v.douyin.com/3NnZSIEzcK0/) --- -## 📊 Benchmark Performance +## 📊 基准测试性能 -**Performance of the Mano series models in multiple benchmarks:** +**Mano系列模型在多项基准测试上的表现：** ### 1. GUI Grounding
-📊 Expand Evaluation Data +📊 展开评测数据
![GUI Agent Grounding Benchmark](./pics/GUI_Agent_Grounding_Benchmark.png) @@ -131,7 +136,7 @@ This video demonstrates the combined capability of [Mano-AFK](#-applications) an ### 2. BUA & CUA
-📊 Expand Evaluation Data +📊 展开评测数据 #### [OSWorld](https://os-world.github.io/) - Specialized Models @@ -150,7 +155,7 @@ This video demonstrates the combined capability of [Mano-AFK](#-applications) an ### 3. Perception & Cognition
-📊 Expand Evaluation Data +📊 展开评测数据 #### Video-SME-2 @@ -652,7 +657,7 @@ This video demonstrates the combined capability of [Mano-AFK](#-applications) an - SalECI + SalECI SSM 0.720 0.599 @@ -706,7 +711,7 @@ This video demonstrates the combined capability of [Mano-AFK](#-applications) an 0.769 0.695 1.735 - 0.868 + 0.868 @@ -813,14 +818,14 @@ This video demonstrates the combined capability of [Mano-AFK](#-applications) an ### 4. Pruning
-📊 Expand Evaluation Data +📊 展开评测数据 #### Online-Mind2Web -**Comparison of Task Execution Success Rate (SR) on Online-Mind2Web Benchmark** -_Avg. Tokens/img_ represents the average visual token retention rate per image; lower values indicate more aggressive pruning. +**对比Online-Mind2Web基准测试中任务执行成功率（SR）** +_Avg. Tokens/img_ 表示每张图片的平均视觉 token 保留率；值越低表示剪枝越激进。 -**GSPruning** is a novel token pruning method designed for Vision-Language Models to efficiently process high-resolution web interfaces by preserving global spatial structure through anchor points and identifying semantic outliers for critical UI elements. It achieves 2-3× throughput speedup with minimal performance loss, enabling more efficient autonomous web agents. +**GSPruning** 是一种新型的token剪枝方法，专为视觉语言模型设计，通过保留全局空间锚点维持网页结构骨架，并识别语义异常值来捕获关键UI元素，从而高效处理高分辨率网页界面。该方法在性能损失极小的情况下实现了2-3倍的吞吐量提升，为构建高效的自主网页智能体设立了新的技术标杆。 @@ -853,25 +858,25 @@ _Avg. Tokens/img_ represents the average visual token retention rate per image; - + - + - + - + @@ -927,7 +932,7 @@ _Avg. Tokens/img_ represents the average visual token retention rate per image; ### 5. Context Learning
-📊 Expand Evaluation Data +📊 展开评测数据 #### [CL Bench](https://github.com/Tencent-Hunyuan/CL-bench) @@ -935,78 +940,78 @@ _Avg. Tokens/img_ represents the average visual token retention rate per image;
-### 6. Mano-P Local · Cloud · Qwen3-VL Task Execution +### 6. Mano-P 端侧 · 云端 · Qwen3-VL 任务执行对比
-📊 Expand Evaluation Data +📊 展开评测数据 -Comparison of four inference configurations on 100 real-machine macOS GUI tasks (MacBook Pro · Apple M5 · 16GB): +在 100 道真机 macOS GUI 任务上(MacBook Pro · Apple M5 · 16GB)对比四种推理配置: -**Overall Metrics** +**总体指标** -| Metric | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | -| ----------------- | :-------------: | :------------------: | :-------------: | :-----------: | -| Pass rate | 83.0% | 47.0% | 56.0% | 39.0% | -| Avg time per step | 9.3s | 8.0s | 7.9s | 10.2s | +| 指标 | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | +| ------------- | :-------------: | :------------------: | :-------------: | :-----------: | +| 通过率 | 83.0% | 47.0% | 56.0% | 39.0% | +| 平均每步耗时 | 9.3s | 8.0s | 7.9s | 10.2s | -**Difficulty Tiers** +**按难度分层** -| Tier | Tasks | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | -| ---- | :---: | :-------------: | :------------------: | :-------------: | :-----------: | -| A | 25 | 23/25 (92%) | 21/25 (84%) | 22/25 (88%) | 18/25 (72%) | -| B | 45 | 37/45 (82%) | 18/45 (40%) | 24/45 (53%) | 14/45 (31%) | -| C | 30 | 23/30 (77%) | 7/30 (23%) | 10/30 (33%) | 7/30 (23%) | +| Tier | 任务数 | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | +| ---- | :----: | :-------------: | :------------------: | :-------------: | :-----------: | +| A | 25 | 23/25 (92%) | 21/25 (84%) | 22/25 (88%) | 18/25 (72%) | +| B | 45 | 37/45 (82%) | 18/45 (40%) | 24/45 (53%) | 14/45 (31%) | +| C | 30 | 23/30 (77%) | 7/30 (23%) | 10/30 (33%) | 7/30 (23%) | -**Per Category** +**按类别** -| Category | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | -| ----------------------------- | :-------------: | :------------------: | :-------------: | :-----------: | -| Browser / Web (31) | 28 (90%) | 23 (74%) | 21 (68%) | 18 (58%) | -| Fuzzy descriptions (10) | 8 (80%) | 3 (30%) | 3 (30%) | 3 (30%) | -| File management (7) | 5 (71%) | 3 (43%) | 3 (43%) | 4 (57%) | -| WeChat (6) | 5 (83%) | 2 (33%) | 4 (67%) | 2 (33%) | -| WeCom / Feishu / DingTalk (6) | 6 (100%) | 2 (33%) | 5 (83%) | 3 (50%) | -| System settings (6) | 3 (50%) | 3 (50%) | 5 (83%) | 3 (50%) | -| WPS / Office (5) | 5 (100%) | 0 (0%) | 2 (40%) | 0 (0%) | -| No open hint (5) | 4 (80%) | 1 (20%) | 3 (60%) | 2 (40%) | -| Notes / Reminders (4) | 4 (100%) | 2 (50%) | 2 (50%) | 0 (0%) | -| System utilities (3) | 3 (100%) | 3 (100%) | 3 (100%) | 1 (33%) | -| Long chains (10) | 8 (80%) | 3 (30%) | 3 (30%) | 2 (20%) | -| Cross-app (5) | 3 (60%) | 0 (0%) | 1 (20%) | 0 (0%) | +| 类别 | Mano-CUA(Cloud) | Mano-CUA-1.0-4B | Mano-CUA-2.0-4B | Qwen3-VL-Plus | +| ------------------ | :-------------: | :------------------: | :-------------: | :-----------: | +| 浏览器/网页 (31) | 28 (90%) | 23 (74%) | 21 (68%) | 18 (58%) | +| 模糊描述 (10) | 8 (80%) | 3 (30%) | 3 (30%) | 3 (30%) | +| 文件管理 (7) | 5 (71%) | 3 (43%) | 3 (43%) | 4 (57%) | +| 微信 (6) | 5 (83%) | 2 (33%) | 4 (67%) | 2 (33%) | +| 企微/飞书/钉钉 (6) | 6 (100%) | 2 (33%) | 5 (83%) | 3 (50%) | +| 系统设置 (6) | 3 (50%) | 3 (50%) | 5 (83%) | 3 (50%) | +| WPS/Office (5) | 5 (100%) | 0 (0%) | 2 (40%) | 0 (0%) | +| 无 open 指令 (5) | 4 (80%) | 1 (20%) | 3 (60%) | 2 (40%) | +| 备忘/提醒 (4) | 4 (100%) | 2 (50%) | 2 (50%) | 0 (0%) | +| 系统工具 (3) | 3 (100%) | 3 (100%) | 3 (100%) | 1 (33%) | +| 长链路 (10) | 8 (80%) | 3 (30%) | 3 (30%) | 2 (20%) | +| 跨应用 (5) | 3 (60%) | 0 (0%) | 1 (20%) | 0 (0%) | **Mano-CUA 1.1 + Bash Tool** -With shell tools enabled in Mano-CUA 1.1, cloud mode can use system commands to handle tasks that pure GUI struggles with: +云端模式在 Mano-CUA 1.1 中接入 shell 工具后,可借助系统命令处理纯 GUI 难以完成的任务: -| ID | Task | GUI Only | +Bash | Method | -| --- | ----------------------- | :------: | :---: | --------------------------- | -| 55 | Finder: tag red label | Fail | Pass | GUI (right-click + tag) | -| 57 | Volume to 50% | Fail | Pass | `osascript 'set volume'` | -| 58 | Max mouse pointer size | Fail | Pass | `defaults write` | -| 82 | Calculator → TextEdit | Fail | Pass | GUI (Calculator + TextEdit) | -| 86 | Translate & save to txt | Fail | Pass | `echo` to file | -| 99 | Rotate image | Fail | Pass | `sips -r 90` | +| ID | 任务 | 纯 GUI | +Bash | 方法 | +| --- | ------------------- | :----: | :---: | -------------------------- | +| 55 | Finder 标记红色标签 | Fail | Pass | GUI(右键 + 标签) | +| 57 | 音量调至 50% | Fail | Pass | `osascript 'set volume'` | +| 58 | 鼠标指针最大 | Fail | Pass | `defaults write` | +| 82 | 计算器→TextEdit | Fail | Pass | GUI(Calculator + TextEdit) | +| 86 | 翻译保存 txt | Fail | Pass | `echo` to file | +| 99 | 旋转图片 | Fail | Pass | `sips -r 90` | -Mano-CUA 1.1+Bash pass rate: **90/100 = 90%** (+7; 2 of those were false negatives caused by a system proxy misconfiguration in the previous run). +Mano-CUA 1.1+Bash 通过率提升至 **90/100 = 90%**(+7,其中 2 道为上轮系统代理误配导致的假阴性）。 -**Why Local Matters** +**端侧的核心价值** -- ✅ **Faster per step**: 7.9s vs Mano-CUA 1.1 9.3s vs Qwen 10.2s -- ✅ **Fully local**: zero outbound traffic for screenshots or task descriptions, no network dependency -- ✅ **Lightweight deployment**: ~6.4 GB memory footprint, runs on a MacBook -- ✅ **Small model beats large model**: Mano-CUA-4B (56%) significantly outperforms the cloud-based general-purpose VL model Qwen3-VL-Plus (39%) as a fully local 4B model, demonstrating the value of GUI-specialized fine-tuning +- ✅ **更快推理**:每步 7.9s vs Mano-CUA 1.1 9.3s vs Qwen 10.2s +- ✅ **完全本地**:截图与任务描述零外发,无网络依赖 +- ✅ **轻量部署**:~6.4 GB 内存即可在 MacBook 上运行 +- ✅ **小模型胜大模型**:Mano-CUA-4B(56%)以全本地 4B 模型大幅超越云端通用 VL 大模型 Qwen3-VL-Plus(39%),体现 GUI 专用微调的价值 -**Current Gap** +**当前差距** -The 4B on-device model reaches 56% vs Mano-CUA 1.1's 83%. The gap concentrates in fuzzy descriptions, cross-app workflows, and deep office-suite operations — these are the explicit directions for the next iteration. Qwen3-VL-Plus, as a cloud-based general-purpose VL model, only reaches 39%, primarily limited by Chinese input focus issues, poor adaptation to non-browser apps, and step-limit truncation — showing that general VL capability does not equal GUI Agent capability. Mano-CUA 1.1+Bash mode pushes the pass rate to **90%** via `osascript` / `sips` / `defaults write`, demonstrating the ceiling of a hybrid GUI + Shell strategy. +端侧 4B 通过率 56% vs 云端 Mano-CUA 1.1 83%,主要差距在模糊描述、跨应用工作流、办公套件深度操作——这也是后续迭代的明确方向。Qwen3-VL-Plus 作为云端通用 VL 大模型仅 39%,主要受限于中文输入焦点问题、非浏览器 app 适配不足以及步数超限,说明通用 VL 能力不等于 GUI Agent 能力。Mano-CUA 1.1+Bash 模式通过 `osascript`/`sips`/`defaults write` 等命令将通过率推至 **90%**,展示了 GUI + Shell 混合策略的上限。 -**Test Configuration** +**测试配置** -- Hardware: MacBook Pro · Apple M5 · 16GB RAM -- Cloud model: Claude Sonnet 4.5 (via `mano.mininglamp.com`) -- Local model: Mano-CUA-4B (W8A16, MLX) -- Qwen model: Qwen3-VL-Plus (via `llm-gateway.mlamp.cn`) -- Task set: 100 tasks covering browser/web, app operations, long chains, cross-app, fuzzy descriptions +- 硬件:MacBook Pro · Apple M5 · 16GB RAM +- 云端模型:Claude Sonnet 4.5(via `mano.mininglamp.com`) +- 端侧模型:Mano-CUA-4B(W8A16,MLX) +- Qwen 模型:Qwen3-VL-Plus(via `llm-gateway.mlamp.cn`) +- 任务集:100 道,覆盖浏览器/网页、App 操作、长链路、跨应用、模糊描述
@@ -1014,233 +1019,231 @@ The 4B on-device model reaches 56% vs Mano-CUA 1.1's 83%. The gap concentrates i ## 🔧 Skills -**Mano-Skill** is a desktop GUI automation tool based on the Mano model, driving cross-platform graphical interface operations through natural language. We provide two different usage forms for the same core capability to adapt to different usage scenarios and user groups. +**Mano-Skill** 是基于 Mano 模型的桌面 GUI 自动化工具，通过自然语言驱动跨平台图形界面操作。同一个核心能力，我们提供了两种不同的使用形式，以适配不同的使用场景和用户群体。 --- -### 📦 Core Capability Overview +### 📦 核心能力概述 -#### Feature Highlights +#### 功能特性 -- **Natural Language Driven**: Users describe tasks in natural language, and the system automatically executes GUI operations -- **Flexible Inference Modes**: - - **Local Mode**: Models run locally, data stays on device, fast response - - Run directly on Mac mini/MacBook (M4 chip or above, 32GB+ RAM) - - Or use Mano-P computing stick (via USB 4.0 connection) - - **Cloud Mode**: Without local model configuration, uses cloud API service (`mano.mininglamp.com`) - - System automatically detects local model configuration and seamlessly switches inference modes -- **Comprehensive Interaction Support**: Click, type, hotkey, scroll, drag, mouse movement, screenshot, wait, app launch, URL navigation -- **Cross-Platform Support**: macOS (stable), Windows, Linux (Beta) +- **自然语言驱动**：用户通过自然语言描述任务，系统自动执行GUI操作 +- **灵活的推理模式**： + - **本地模式**：模型在本地运行，数据不出设备，快速响应 - 直接在 Mac mini/MacBook（M4 芯片及以上，32GB+ 内存）上运行 - 或使用 Mano-P 算力棒（通过 USB 4.0 连接） + - **云端模式**：未配置本地模型时，使用云端 API 服务 (`mano.mininglamp.com`) + - 系统自动检测本地模型配置，无缝切换推理模式 +- **全方位交互支持**：点击、输入、热键、滚动、拖拽、鼠标移动、截图、等待、应用启动、URL跳转 +- **跨平台支持**：macOS（稳定）、Windows、Linux（Beta） -#### How It Works +#### 工作原理 -**Cloud Mode (Default)** +**云端模式（默认）** -1. Capture current screen screenshot -2. Send screenshot and task description to cloud vision model (`mano.mininglamp.com`) -3. Cloud model analyzes and returns next action instruction -4. Local client executes operation (click, type, etc.) -5. Loop execution until task completion +1. 捕获当前屏幕截图 +2. 发送截图和任务描述到云端视觉模型 (`mano.mininglamp.com`) +3. 云端模型分析并返回下一步操作指令 +4. 本地客户端执行操作（点击、输入等） +5. 循环执行直到任务完成 -**Local Mode** +**本地模式** -1. Capture current screen screenshot -2. Run Mano-P model on local device (Mac mini/MacBook) or computing stick for inference -3. Local model analyzes and returns next action instruction -4. Client executes operation (click, type, etc.) -5. Loop execution until task completion +1. 捕获当前屏幕截图 +2. 在本地设备（Mac mini/MacBook）或算力棒上运行 Mano-P 模型进行推理 +3. 本地模型分析并返回下一步操作指令 +4. 客户端执行操作（点击、输入等） +5. 循环执行直到任务完成 -#### Data Privacy & Security +#### 数据隐私与安全 -**Cloud Mode:** +**云端模式：** -- ⚠️ **Data Sent**: Screenshots and task descriptions sent to `mano.mininglamp.com` for real-time visual analysis -- ✅ **Data Not Sent**: Does not access or transmit local files, clipboard contents, system credentials -- ⚠️ **Privacy Note**: Avoid displaying sensitive documents, chat logs, or credential information on screen when running tasks +- ⚠️ **发送数据**：截图和任务描述发送到 `mano.mininglamp.com` 进行实时视觉分析 +- ✅ **不发送数据**：不访问或传输本地文件、剪贴板内容、系统凭证 +- ⚠️ **隐私提示**：运行任务时避免在屏幕上显示敏感文档、聊天记录或凭证信息 -**Local Mode (Mac mini/MacBook or Computing Stick):** +**本地模式（Mac mini/MacBook 或算力棒）：** -- ✅ **Fully Local Processing**: All data processing is completed locally, screenshots and task descriptions never leave the device -- ✅ **Data Stays on Device**: Does not access or transmit any data to external servers -- ✅ **Maximum Privacy Protection**: Suitable for handling sensitive information and high-security scenarios +- ✅ **完全本地运行**：所有数据处理在本地完成，截图和任务描述完全不上云 +- ✅ **数据不出设备**：不访问或传输任何数据到外部服务器 +- ✅ **最高隐私保护**：适合处理敏感信息和高安全要求场景 -**General Assurance:** +**通用保障：** -- ✅ **Open Source Auditable**: Complete source code publicly available for review +- ✅ **开源可审计**：完整源代码公开，可供审查 --- -### 🔧 Two Usage Forms +### 🔧 两种使用形式 -> If you want to use Mano-P directly to accomplish GUI automation tasks, here are two different usage forms. Choose the one that best fits your use case. +> 如果您想直接使用 Mano-P 完成 GUI 自动化任务，这里提供了两种不同的使用方式，根据您的使用场景选择最适合的形式。 -#### 1️⃣ mano-cua (CLI Command-Line Tool) +#### 1️⃣ mano-cua（CLI 命令行工具） -**Use Case**: **For human users** — developers and advanced users invoking mano-cua directly in a terminal to quickly execute one-off or scripted GUI automation tasks +**适用场景**：**面向人类用户**——开发者、高级用户在终端手动调用，快速完成一次性或脚本化的 GUI 自动化任务 -**Installation**: +**安装方式**： ```bash -# Install via Homebrew +# 通过 Homebrew 安装 brew tap Mininglamp-AI/tap brew install mano-cua ``` -The installation process will automatically: +安装过程会自动完成： -- Create an isolated Python 3.13 virtual environment -- Install required dependencies (including Tkinter GUI library) -- Configure the executable command to system PATH +- 创建独立的 Python 3.13 虚拟环境 +- 安装所需依赖（包括 Tkinter 图形界面库） +- 配置可执行命令到系统路径 -**Usage**: +**使用方式**： ```bash -# Cloud mode (default — no extra setup required) -mano-cua run "Open WeChat and tell FTY the meeting is postponed" -mano-cua run "Search for AI news on Xiaohongshu and display the first post" +# 云端模式（默认，无需额外配置） +mano-cua run "打开微信并告诉FTY会议延期" +mano-cua run "在小红书搜索AI新闻并展示第一条帖子" -# Stop current task +# 停止当前任务 mano-cua stop ``` -mano-cua ships with two inference modes: **cloud** and **local**. Cloud is the default; to run Mano-P locally on macOS Apple Silicon, use the `--local` flag: +mano-cua 内置**云端**与**本地**两种推理模式。默认走云端（由 Mininglamp 托管的推理服务 `mano.mininglamp.com`）；若要在本地运行 Mano-P（macOS Apple Silicon），使用 `--local` 标志： ```bash -# First-time local setup: verify env / install SDK / pull the local model +# 首次本地使用：检查环境、安装 SDK、拉取本地模型 mano-cua check mano-cua install-sdk mano-cua install-model -# Run a task in local mode -mano-cua run "Open Safari and search for Python" --local -mano-cua run "Type hello in the search box" --local --url "https://www.baidu.com" --minimize --max-steps 15 +# 以本地模式运行任务 +mano-cua run "打开 Safari 并搜索 Python" --local +mano-cua run "在搜索框中输入 hello" --local --url "https://www.baidu.com" --minimize --max-steps 15 ``` -In local mode, Mano-P runs on-device via MLX — screenshots and task descriptions make zero network calls for inference. +本地模式下，Mano-P 基于 MLX 直接在设备上推理，所有截图与任务描述不经过任何网络调用。 -**Features**: +**特点**： -- ✅ Command-line interface, quick invocation -- ✅ Virtual environment isolation, no system Python pollution -- ✅ Suitable for script integration and batch processing -- ✅ Can be embedded in shell scripts -- ✅ Cloud / local inference modes — `--local` switches to on-device in one flag +- ✅ 命令行界面，快速调用 +- ✅ 虚拟环境隔离，不污染系统 Python +- ✅ 适合脚本集成和批处理 +- ✅ 可在 shell 脚本中嵌入使用 +- ✅ 云端 / 本地两种推理模式，`--local` 一键切换到设备本地 -**Installation & Distribution**: +**安装与分发**： -- **Homebrew Tap**: [github.com/Mininglamp-AI/homebrew-tap](https://github.com/Mininglamp-AI/homebrew-tap) +- **Homebrew Tap**：[github.com/Mininglamp-AI/homebrew-tap](https://github.com/Mininglamp-AI/homebrew-tap) --- -#### 2️⃣ mano-skill (ClawHub Skill Form) +#### 2️⃣ mano-skill（ClawHub Skill 形式） -**Use Case**: **For AI agents** — Claude Code, OpenClaw and similar agents autonomously invoke GUI automation capabilities mid-reasoning to complete user tasks, no manual command execution needed +**适用场景**：**面向 AI Agent**——由 Claude Code、OpenClaw 等 agent 在推理链路中自主调用 GUI 自动化能力完成用户任务，无需用户手动触发命令 -**Installation**: +**安装方式**： -**Option 1: Install via Claude Code** +**方案一：通过 Claude Code 安装** -In Claude Code, skills exist as "commands". Installation steps: +在 Claude Code 中，skills 以"命令"(commands) 的形式存在。安装步骤： -1. Download the skill zip package from [ClawHub](https://clawhub.ai/hanningwang/mano-cua) -2. After extraction, copy files to Claude Code's commands directory -3. Restart Claude Code or in a new session, the skill will be automatically available +1. 从 [ClawHub](https://clawhub.ai/hanningwang/mano-cua) 下载 skill zip 包 +2. 解压后将文件复制到 Claude Code 的 commands 目录： +3. 重启 Claude Code 或在新会话中，skill 将自动可用 -**Option 2: Install via ClawHub CLI (Recommended)** +**方案二：通过 ClawHub CLI 安装（推荐）** -Use the ClawHub CLI tool for one-click installation and skill management: +使用 ClawHub CLI 工具可以一键安装和管理 skills： ```bash -# Install skill +# 安装 skill clawhub install mano-cua -# Install specific version +# 安装特定版本 clawhub install mano-cua --version 1.0.0 -# Update skill to latest version +# 更新 skill 到最新版本 clawhub update mano-cua ``` -After installation, start a new Claude Code or OpenClaw session to use. +安装完成后，启动新的 Claude Code 或 OpenClaw 会话即可使用。 -> **Prerequisites**: ClawHub CLI tool must be installed first. See: [OpenClaw Documentation - ClawHub](https://docs.openclaw.ai/tools/clawhub) +> **前置要求**：需要先安装 ClawHub CLI 工具。详见：[OpenClaw 文档 - ClawHub](https://docs.openclaw.ai/tools/clawhub) -**Usage**: +**使用方式**： -When users make requests to AI agents that require GUI operations, the agent will automatically invoke this skill: +当用户向 AI Agent 提出需要 GUI 操作的需求时，Agent 会自动调用此 skill： ``` -User: "Help me open WeChat, find FTY's chat window, and tell him the meeting is postponed to tomorrow" -Agent: [Automatically invokes mano-skill to complete GUI operation] +用户: "帮我打开微信，找到FTY的聊天窗口，告诉他会议延期到明天" +Agent: [自动调用 mano-skill 完成 GUI 操作] ``` -**Features**: +**特点**： -- ✅ Autonomously invoked by AI agents, no manual command execution needed -- ✅ Deeply integrated with agent reasoning capabilities -- ✅ Suitable for complex multi-step task automation -- ✅ ClawHub ecosystem with version management and security scanning +- ✅ AI Agent 自主调用，无需用户手动执行命令 +- ✅ 与 Agent 的推理能力深度集成 +- ✅ 适合复杂的多步骤任务自动化 +- ✅ ClawHub 生态，支持版本管理和安全扫描 -**Project Resources**: +**项目资源**： -- **Source Code**: [github.com/Mininglamp-AI/mano-skill](https://github.com/Mininglamp-AI/mano-skill) -- **ClawHub Home**: [clawhub.ai/hanningwang/mano-cua](https://clawhub.ai/hanningwang/mano-cua) -- **Version**: v1.0.0 -- **License**: MIT +- **源代码**：[https://github.com/Mininglamp-AI/mano-skill](https://github.com/Mininglamp-AI/mano-skill) +- **ClawHub 主页**：[clawhub.ai/hanningwang/mano-cua](https://clawhub.ai/hanningwang/mano-cua) +- **版本**：v1.0.0 +- **许可证**：MIT --- -### ⚙️ Permission Requirements (Common to All Forms) +### ⚙️ 权限要求（所有形式通用） -- **Screen Recording Permission** -- **Accessibility Permission** (keyboard/mouse control) -- Grant permissions in **System Preferences → Privacy & Security** +- **屏幕录制权限**（Screen Recording） +- **辅助功能权限**（Accessibility - 键盘/鼠标控制） +- 在 **系统偏好设置 → 隐私与安全** 中授予权限 -### 🔒 Security Constraints (Common to All Forms) +### 🔒 安全约束（所有形式通用） -- Sensitive or potentially dangerous operations require user confirmation before execution -- Users can stop tasks at any time -- Only one task can run on each device simultaneously -- Only supports primary display (multi-display environment) +- 敏感或潜在危险的操作需要用户确认才能执行 +- 用户可随时停止任务 +- 每台设备同时只能运行一个任务 +- 仅支持主显示器（多显示器环境） -### 📊 Status Panel +### 📊 状态面板 -When a task is running, a small status panel appears in the top-right corner of the screen to: +任务运行时，屏幕右上角会显示一个小型状态面板，用于： -- Display real-time task status and progress -- Provide task management functions (pause/stop) -- Remind users that an automation task is running to avoid accidental interference +- 实时显示当前任务状态和进度 +- 提供任务管理功能（暂停/停止） +- 提醒用户当前有自动化任务正在运行，避免误操作 -### 🔔 Platform Compatibility Note +### 🔔 平台兼容性说明 -**Beta Version Notice**: Mano-Skill is currently in Beta testing phase. +**Beta 版本提示**：Mano-Skill 当前处于 Beta 测试阶段。 -- **macOS**: ✅ Preferred and most thoroughly tested platform, stable and ready for use -- **Windows** and **Linux**: ⚠️ Platform adaptations not yet fully completed, minor issues may occur +- **macOS**：✅ 首选和测试最充分的平台，稳定可用 +- **Windows** 和 **Linux**：⚠️ 平台适配尚未完全完成，可能存在小问题 -We are continuously improving cross-platform compatibility. Feedback is welcome. +我们正在持续改进跨平台兼容性，欢迎反馈使用体验。 --- -## 🤖 Models +## 🤖 模型 -> If you want to integrate Mano-P's model capabilities into your own applications, this section provides performance metrics and usage guidelines. +> 如果您想在自己的应用中集成 Mano-P 的模型能力，这里提供了模型的性能指标和使用指南。 -### Performance Evaluation +### 性能评测 -The table below presents actual inference benchmark results of Mano-CUA-4B running on Apple M5 Pro with the Cider inference SDK. Using W8A16 (MLX's native weight-only quantization path) as the baseline — the same reference convention adopted in [Cider's quantization benchmark](#-inference-sdk) — enabling Cider's W8A8 activation quantization reduces prefill time from 2.839s to 2.519s on the same input, a **~12.7% prefill speedup**. For more data, refer to the [⚡ Inference SDK](#-inference-sdk) section below. +下表展示了 Mano-CUA-4B 在 Apple M5 Pro + Cider 推理 SDK 组合下的实际性能基准。以 W8A16（MLX 原生仅权重量化路径，也是 [Cider quantization benchmark](#-推理加速-sdk) 采用的参照系之一）作为 baseline，启用 Cider 的 W8A8 激活量化后，同一输入下 prefill 时间从 2.839s 降到 2.519s，**prefill 约加速 12.7%**。更多数据参考下文 [⚡ 推理加速 SDK](#-推理加速-sdk) 章节。
0.310
FlashVLM [4]FlashVLM 12.55% 17.01 0.343
Compressor-VLA [11]Compressor-VLA 13.33% 16.92 0.293
HiPrune [16]HiPrune 25.09% 16.67 0.333
PDrop [33]PDrop 41.47% 10.43 0.330
- - - - - - - - + + + + + + + + @@ -1262,9 +1265,9 @@ The table below presents actual inference benchmark results of Mano-CUA-4B runni
ModelChipBandwidthFrameworkContext LengthQuantizationPrefill Time
(s)		Decode Speed
(tokens/s)		模型芯片带宽框架上下文长度量化Prefill 时间
(s)		Decode 速度
(tokens/s)
-**Model Download:** +**模型下载：** -| Platform | Base Model | MLX 8-bit Quantized | +| 平台 | 基础模型 | MLX 8-bit 量化 | | ---------------- | ---------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | 🤗 Hugging Face | [Mano-CUA-2.0-4B](https://huggingface.co/Mininglamp-2718/Mano-CUA-2.0-4B) | [Mano-CUA-2.0-4B-MLX-8bit](https://huggingface.co/Mininglamp-2718/Mano-CUA-2.0-4B-MLX-8bit) | | 🪄 ModelScope CN | [Mano-CUA-2.0-4B](https://www.modelscope.cn/models/Mininglamp2718/Mano-CUA-2.0-4B) | [Mano-CUA-2.0-4B-MLX-8bit](https://www.modelscope.cn/models/Mininglamp2718/Mano-CUA-2.0-4B-MLX-8bit) | @@ -1272,52 +1275,52 @@ The table below presents actual inference benchmark results of Mano-CUA-4B runni --- -## ⚡ Inference SDK +## ⚡ 推理加速 SDK -### Overview +### 概述 -**Cider** is an inference acceleration SDK developed on top of MLX for macOS. It provides online activation quantization operators absent in MLX, with custom int-matmul kernels built as MLX custom primitives supporting full lazy evaluation. It also includes service-side extensions and non-intrusive compatibility patches for `mlx_vlm` (validated on `mlx_vlm 0.4.3`), including fixes for Qwen3-VL multi-image inference issues related to RoPE position handling and chunked prefill. +**Cider** 是面向 macOS 的推理加速 SDK，基于 Apple MLX 构建。它提供 MLX 原生未包含的在线激活量化算子，以自定义 INT matmul kernel 的形式作为 MLX 自定义原语对外暴露，支持完整的 lazy evaluation。Cider 同时提供服务端扩展以及对 `mlx_vlm` 的非侵入式兼容补丁（在 `mlx_vlm 0.4.3` 上验证），修复了 Qwen3-VL 多图推理中 RoPE 位置处理与 chunked prefill 相关的若干问题。 -### Conditional Compilation (M4 / M5) +### 条件编译（M4 / M5） -Cider uses **conditional compilation**: the INT8 TensorOps C++ extension is only built on Apple M5+. +Cider 采用**条件编译**策略：INT8 TensorOps C++ 扩展仅在 Apple M5 及以上芯片构建。 -| Chip | `pip install -e .` behavior | `import cider` behavior | -| ---------------- | --------------------------------------------- | -------------------------------------------------------------- | -| **M5+** | Full build (CMake + Metal kernels) | All features available | -| **M4 and below** | Skips C++ build, installs pure-Python package | `is_available()` → False, `convert_model()` is a warning no-op | +| 芯片 | `pip install -e .` 行为 | `import cider` 行为 | +| ------------- | --------------------------------- | -------------------------------------------------------------- | +| **M5+** | 完整构建（CMake + Metal kernels） | 所有功能可用 | +| **M4 及以下** | 跳过 C++ 构建，仅安装纯 Python 包 | `is_available()` → False，`convert_model()` 为空操作并提示警告 | -**Override via environment variable:** +**通过环境变量覆盖：** ```bash -CIDER_FORCE_BUILD=1 pip install -e . # Force build (e.g., CI) -CIDER_FORCE_BUILD=0 pip install -e . # Force skip +CIDER_FORCE_BUILD=1 pip install -e . # 强制构建（例如 CI 场景） +CIDER_FORCE_BUILD=0 pip install -e . # 强制跳过 ``` ### Modes -| Mode | Weights | Activations | Compute Path | Status | -| -------- | ------------------- | -------------- | ------------------ | -------------- | -| **W8A8** | INT8 symmetric | INT8 per-token | TensorOps matmul2d | ✅ Implemented | -| **W4A8** | INT4 packed (uint8) | INT8 per-token | Unpack → TensorOps | ✅ Implemented | -| W4A16 | — | — | MLX built-in | Baseline | -| W8A16 | — | — | MLX built-in | Baseline | +| Mode | Weights | Activations | Compute Path | Status | +| -------- | ------------------- | -------------- | ------------------ | --------- | +| **W8A8** | INT8 symmetric | INT8 per-token | TensorOps matmul2d | ✅ 已实现 | +| **W4A8** | INT4 packed (uint8) | INT8 per-token | Unpack → TensorOps | ✅ 已实现 | +| W4A16 | — | — | MLX built-in | Baseline | +| W8A16 | — | — | MLX built-in | Baseline | -**W4A16 and W8A16 are already supported by MLX natively** — this SDK provides the missing **W8A8** and **W4A8** modes that MLX does not implement. +**MLX 原生已支持 W4A16 与 W8A16**——Cider 补齐了 MLX 未实现的 **W8A8** 与 **W4A8** 两种模式。 -MLX's quantization is **weight-only**: `QuantizedLinear` dequantizes weights to FP16 and uses FP16 GEMM. While MLX's Steel NAX templates are generic enough to be instantiated with INT8 types (and would achieve identical raw matmul throughput — [see the transparent benchmark](https://github.com/Mininglamp-AI/cider/blob/main/benchmarks/mlx_native/cider_vs_mlx_int8.md)), MLX does not provide the quantization/dequantization pipeline needed for actual W8A8 inference. Cider fills this gap with fused quantize-matmul-dequant primitives, implementing online INT8 activation quantization and INT8 TensorOps-based compute for the supported inference paths. +MLX 的量化是**仅权重量化**：`QuantizedLinear` 将权重反量化为 FP16 后走 FP16 GEMM。虽然 MLX 的 Steel NAX 模板原则上可以实例化为 INT8 类型（实测原始 matmul 吞吐量相当——参见[透明基准对比](https://github.com/Mininglamp-AI/cider/blob/main/benchmarks/mlx_native/cider_vs_mlx_int8.md)），但 MLX 并未提供真正的 W8A8 推理所需的量化/反量化流水线。Cider 通过 fused quantize-matmul-dequant 原语补齐了这一能力，为支持的推理路径提供在线 INT8 激活量化 + INT8 TensorOps 计算。 -#### W8A8 Quantization Granularity +#### W8A8 量化粒度 -| Granularity | Description | Speed | Precision | -| ---------------------- | ---------------------------- | ----------------------- | ---------------------------- | -| **Per-channel** | One scale per output channel | Fastest (1.8x prefill) | Slightly lower | -| **Per-group (gs=128)** | One scale per 128 elements | Fast (1.5x prefill) | Moderate precision retention | -| **Per-group (gs=64)** | One scale per 64 elements | Moderate (1.3x prefill) | Higher precision | +| 粒度 | 描述 | 速度 | 精度 | +| ---------------------- | ----------------------- | -------------------- | ------------ | +| **Per-channel** | 每个输出通道一个 scale | 最快（prefill 1.8x） | 略低 | +| **Per-group (gs=128)** | 每 128 个元素一个 scale | 快（prefill 1.5x） | 精度保留适中 | +| **Per-group (gs=64)** | 每 64 个元素一个 scale | 中等（prefill 1.3x） | 精度保留较高 | -### Performance (Apple M5 Pro) +### Performance（Apple M5 Pro） -**Individual Operator Latency** +**单算子延迟** Shape [N=10240, K=2560] @@ -1339,7 +1342,7 @@ Shape [N=2560, K=10240] | 4096 | 5.37ms | 6.79ms | 8.54ms | 8.04ms | 1.59x | 1.50x | 1.26x | 1.18x | | 8192 | 10.97ms | 12.94ms | 17.28ms | 16.23ms | 1.58x | 1.48x | 1.34x | 1.25x | -**End-to-End VLM** +**端到端 VLM** _Qwen3-VL-2B_ @@ -1357,16 +1360,16 @@ _Qwen3-VL-4B_ | 2393 | 1815 | 1700 | **2028** | 31 | **55** | 52 | | 3455 | 1755 | 1603 | **1881** | 30 | **52** | 49 | -**LLM Quantization: Precision vs. Speed Comparison** +**LLM 量化：精度与速度对比** - - - + + + @@ -1437,107 +1440,107 @@ _Qwen3-VL-4B_
Models Quantization Configurationwikitext2 PPL (↓)Prefill Time (s) (↓)Peak Memory (GB) (↓)wikitext2 PPL（↓）Prefill Time (s)（↓）Peak Memory (GB)（↓）
-- 🔗 Repository: [github.com/Mininglamp-AI/cider](https://github.com/Mininglamp-AI/cider) +- 🔗 仓库地址：[github.com/Mininglamp-AI/cider](https://github.com/Mininglamp-AI/cider) --- -## ⚗️ Approach +## ⚗️ 方法 -> If you are a researcher or wish to train customized GUI Agent models based on your own data, we plan to open-source the complete Mano-Action training methodology and related tools. +> 如果您是研究人员或希望基于自己的数据训练定制化 GUI Agent 模型，我们计划开源完整的 Mano-Action 训练方法和相关工具。 > -> **Release Soon** +> **即将开源** -### Mano-Action Training Methodology +### Mano-Action 训练方法 -Mano-Action is a bidirectional self-reinforcement training framework specifically designed for GUI Grounding. Unlike traditional unidirectional prediction methods, Mano-Action achieves more robust interface understanding through Text↔Action cycle consistency learning, enabling the model to master both "locating elements from descriptions" and "describing given elements" simultaneously. +Mano-Action 是一套专为 GUI Grounding 设计的双向自增强训练框架。不同于传统的单向预测方法，Mano-Action 通过 Text↔Action 循环一致性学习，让模型同时掌握"从描述定位元素"和"描述给定元素"两个方向的能力，从而实现更鲁棒的界面理解。 -#### Core Features +#### 核心特性 -- **Bidirectional Cycle Learning**: Mutual reinforcement between Text → Action and Action → Text -- **Three-Stage Progressive Training**: Supervised Learning → Offline RL → Online RL -- **Closed-Loop Data Generation**: Automatically generate high-quality training data for continuous model improvement -- **Edge Optimization Adaptation**: Includes quantization, pruning, and other edge deployment optimization techniques +- **双向循环学习**：Text → Action 和 Action → Text 的互相增强 +- **三阶段渐进训练**：监督学习 → 离线强化学习 → 在线强化学习 +- **闭环数据生成**：自动生成高质量训练数据，持续提升模型能力 +- **端侧优化适配**：包含量化、剪枝等端侧部署优化技术 -#### Use Cases +#### 适用场景 -- 🎓 **Academic Research**: Explore new approaches to GUI understanding and multimodal interaction -- 🏢 **Enterprise Customization**: Train specialized models based on internal enterprise systems -- 🌐 **Domain Adaptation**: Fine-tune models for specific domains (healthcare, finance, etc.) -- 🔬 **Algorithm Innovation**: Develop new training techniques building on Mano-Action +- 🎓 **学术研究**：探索 GUI 理解和多模态交互新方法 +- 🏢 **企业定制**：基于企业内部系统训练专用模型 +- 🌐 **领域适配**：针对特定领域（医疗、金融等）微调模型 +- 🔬 **算法创新**：在 Mano-Action 基础上开发新的训练技术 --- -## 🌟 Technical Advantages +## 🌟 技术优势 -### Mano-P vs Traditional Solutions CUA Comparison +### Mano-P vs 传统方案 CUA对比 -| Feature | Mano-P | OpenClaw | Manus | Traditional RPA | -| ------------------- | -------------------------------- | ------------------------ | ---------------------------- | ------------------------------- | -| **Model Source** | ✅ Built-in edge model | ⚠️ User-configured | ⚠️ Cloud API calls | ❌ No model (rule-based engine) | -| **Data Security** | ✅ Local execution | ⚠️ LLM/skill cloud calls | ⚠️ Cloud inference | ✅ Can be local | -| **Control Method** | ✅ Pure visual | ⚠️ CDP protocol+CLI | ❌ HTML parsing+CLI | ❌ System API | -| **Use Scenarios** | ✅ All-type GUI (desktop/Web/3D) | ✅ Multi-type apps | ⚠️ Web apps only | ⚠️ Specific systems | -| **Long Task Plan** | ✅ Autonomous planning | ✅ Autonomous planning | ✅ Visual flow orchestration | ❌ Needs preset workflows | -| **Response Speed** | ✅ Instant response | ✅ Local/cloud execution | ⚠️ Cloud latency | ✅ Instant response | -| **Deployment Cost** | ✅ Low-cost entry | ✅ Open source & free | ⚠️ Subscription fee | ✅ Low cost | -| **Robustness** | ✅ UI change adaptive | ✅ LLM adaptive | ⚠️ Limited adaptation | ❌ UI change needs reconfig | +| 特性 | Mano-P | OpenClaw | Manus | 传统RPA | +| -------------- | --------------------------- | -------------------- | ----------------- | ----------------------- | +| **模型来源** | ✅ 内置端侧模型 | ⚠️ 用户自行配置 | ⚠️ 云端API调用 | ❌ 无需模型（规则引擎） | +| **数据安全** | ✅ 本地运行 | ⚠️ LLM/skill调用上云 | ⚠️ 云端推理 | ✅ 可本地 | +| **控制方式** | ✅ 纯视觉交互 | ⚠️ CDP协议+CLI | ❌ HTML解析+CLI | ❌ 系统API | +| **适用场景** | ✅ 全类型GUI（桌面/Web/3D） | ✅ 多类型应用 | ⚠️ 仅Web应用 | ⚠️ 特定系统 | +| **长任务规划** | ✅ 自主规划 | ✅ 自主规划 | ✅ 可视化流程编排 | ❌ 需预设流程 | +| **响应速度** | ✅ 即时响应 | ✅ 本地/云端运行 | ⚠️ 云端延迟 | ✅ 即时响应 | +| **部署成本** | ✅ 低成本起步 | ✅ 开源免费 | ⚠️ 订阅付费 | ✅ 低成本 | +| **鲁棒性** | ✅ UI变化自适应 | ✅ LLM自适应 | ⚠️ 有限适应 | ❌ 界面变化需重配 | -### Core Competitiveness +### 核心竞争力 -1. **Edge Large Model + Flexible Deployment** - - 4B model runs directly on Mac (M4 chip + 32GB RAM) - - Large parameter models (72B) supported via computing stick - - No API key configuration needed, ready out-of-the-box - - Significant advantage over OpenClaw (requires user model configuration) and Manus (cloud calls) +1. **端侧大模型 + 灵活部署** + - 4B模型可直接在Mac上运行（M4芯片+32GB内存） + - 大参数模型（72B）通过算力棒支持 + - 无需配置API密钥，开箱即用 + - 相对OpenClaw（需用户配置模型）和Manus（云端调用）的显著优势 -2. **Universal Visual Understanding** - - Pure visual GUI interaction, not limited to browsers and web apps - - Broader support than OpenClaw (CDP protocol mainly for browsers) and Manus (web apps only) - - Supports desktop software, 3D applications, professional tools, and non-standard GUIs +2. **全场景视觉理解** + - 纯视觉GUI交互，不限于浏览器和Web应用 + - 相对OpenClaw（CDP协议主要针对浏览器）和Manus（仅Web应用）支持更广 + - 支持桌面软件、3D应用、专业工具等非标准GUI -3. **Offline Long-Task Autonomous Planning** - - Fully offline reasoning for complex business processes - - Autonomous decision-making and error correction without internet connection - - Unique advantage over Manus (cloud latency) and traditional RPA (needs preset workflows) +3. **离线长任务自主规划** + - 复杂业务流程的完全离线推理 + - 无需联网即可完成自主决策与纠错 + - 相对Manus（云端延迟）和传统RPA（需预设流程）的独特优势 -4. **Integrated Hardware Deployment** - - Model + computing stick integrated solution, plug-and-play - - Lowers technical barrier compared to OpenClaw (open-source & free but requires self-deployment) - - Multiple deployment forms (direct Mac install / plug-and-play compute stick), rapid launch +4. **一体化硬件部署** + - 模型+算力棒一体化方案，即插即用 + - 相对OpenClaw（开源免费但需自行部署）降低技术门槛 + - 多部署形态（Mac 直装 / 算力棒即插即用），快速上线 --- -## 🔗 Applications +## 🔗 应用 -**Mano-AFK** — an autonomous full-cycle app builder that turns a single natural-language sentence into a deployed, tested, and bug-fixed application. Its E2E testing stage runs **by default with Mano-P as the local backend** (screenshots and task descriptions stay on-device); it can also switch to cloud mode driven by Anthropic's Claude CUA (requires `ANTHROPIC_API_KEY`; third-party service with no local alternative). Mano-AFK is a concrete application scenario of Mano-P in real-world software engineering pipelines. +**Mano-AFK** —— 端到端的自动化应用构建流水线，从一句自然语言描述出发，自主完成需求澄清、架构设计、代码生成、部署、测试与修复，最终交付可运行的应用。E2E 测试环节**默认以 Mano-P 为本地后端**（截图与任务描述不出设备）；也可切换到云端模式，由 Anthropic Claude CUA 驱动（需 `ANTHROPIC_API_KEY`，第三方服务无本地替代）。Mano-AFK 是 Mano-P 在软件工程场景中的典型落地案例。 - GitHub: [github.com/Mininglamp-AI/mano-afk](https://github.com/Mininglamp-AI/mano-afk) - ClawHub: [clawhub.ai/hanningwang/mano-afk](https://clawhub.ai/hanningwang/mano-afk) -**CUA Benchmark — Mano-CUA-4B on MacBook Pro M5 (16GB)** +**CUA Benchmark —— Mano-CUA-4B 运行于 MacBook Pro M5（16GB）** -The suite evaluates 100 tasks across 5 web applications that were themselves built autonomously by Mano-AFK: **TripSplit** (expense splitting), **md-wechat** (Markdown → WeChat formatter), **OMS** (order management), **Family Ledger** (household bookkeeping), and **Life Dashboard** (personal widgets). Each app ships in two variants — a **golden** build (bug-free, expected verdict PASS, 76 tasks) and a **buggy** build with specific UI/logic defects injected (expected verdict FAIL, 24 tasks). Accuracy is defined as the share of tasks where the judge's verdict matches the expected label; each project contributes 15–16 golden tasks and 4–5 bug-injection tasks. +测试集包含 100 条任务，覆盖 5 个由 Mano-AFK 自身自动构建的 Web 应用：**TripSplit**（差旅分账）、**md-wechat**（Markdown → 微信排版）、**OMS**（订单管理）、**Family Ledger**（家庭记账）、**Life Dashboard**（个人仪表盘）。每个应用都有两个版本：**golden 版**为无 bug 构建（预期 judge 判定 PASS，共 76 条任务）；**buggy 版**则预先注入了 UI 或逻辑缺陷（预期 judge 判定 FAIL，共 24 条任务）。准确率定义为"judge 判定与预期结果一致"的任务占比；每个项目贡献 15–16 条 golden 任务 + 4–5 条 bug 注入任务。 -| Configuration | Accuracy | Avg Steps | Prefill Speed | Avg Tokens/Step | -| ------------------------------------------------------ | --------- | --------- | ------------- | --------------- | -| W8A16 | **58.0%** | 6.1 | ~1,253 tok/s | 3,389 | -| W8A8 ([Cider](https://github.com/Mininglamp-AI/cider)) | **54.0%** | 6.93 | ~1,453 tok/s | 3,104 | +| 配置 | 准确率 | 平均步数 | Prefill 速度 | 平均 Tokens/步 | +| ------------------------------------------------------ | --------- | -------- | ------------ | -------------- | +| W8A16 | **58.0%** | 6.1 | ~1,253 tok/s | 3,389 | +| W8A8 ([Cider](https://github.com/Mininglamp-AI/cider)) | **54.0%** | 6.93 | ~1,453 tok/s | 3,104 | -Metrics: _Accuracy_ — judge verdict matches expected PASS/FAIL; _Steps_ — actions taken per task; _Prefill Speed_ — token throughput during the prefill phase; _Tokens/Step_ — prompt + generation tokens combined. +指标说明：_准确率_ —— judge 判定与预期 PASS/FAIL 一致的任务占比；_步数_ —— 单个任务中 agent 执行的动作数；_Prefill 速度_ —— prefill 阶段的 token 吞吐；_Tokens/步_ —— prompt 与 generation 的 token 合计。 -> **Note on W8A8 on this hardware:** The W8A8 row runs via [Cider](https://github.com/Mininglamp-AI/cider) for INT8 activation quantization. W8A8 accelerates prefill through INT8 TensorOps, but it must hold both the original and the INT8 weights in memory simultaneously — roughly doubling weight footprint. On a 16 GB unified-memory device, the added pressure can trigger swapping that offsets the prefill gain, which is why W8A8 here lands slightly behind W8A16 on accuracy. At least 4 GB of free memory beyond the model size is recommended to see W8A8's full benefit. +> **关于此硬件上的 W8A8：** W8A8 行通过 [Cider](https://github.com/Mininglamp-AI/cider) 实现 INT8 激活量化。W8A8 通过 INT8 TensorOps 加速 prefill，但需要同时保留原始权重与 INT8 权重，权重内存近似翻倍。在 16 GB 统一内存设备上，额外的内存压力可能触发换页，抵消 prefill 的加速收益，因此本测试中 W8A8 的准确率反而略差于 W8A16。建议预留超出模型体积 4 GB 以上的空闲内存，才能发挥 W8A8 的完整优势。 -- [Benchmark methodology](https://github.com/Mininglamp-AI/mano-afk/tree/master/benchmark) — full test design, task schema, and bug catalog +- [Benchmark methodology](https://github.com/Mininglamp-AI/mano-afk/tree/master/benchmark) —— 完整测试设计、任务 schema 与 bug 清单 --- -## 📄 Technical Papers & Citation +## 📄 技术论文与引用 -### Related Papers +### 相关论文 -Mano-P is based on the following research work: +Mano-P基于以下研究工作： -**1. Mano Series Model Foundation Paper** +**1. Mano系列模型基础论文** ```bibtex @article{mano-2025, @@ -1549,7 +1552,7 @@ Mano-P is based on the following research work: } ``` -**2. WebRetriever Benchmark** +**2. WebRetriever基准测试** ```bibtex @article{webretriever-2026, @@ -1561,162 +1564,162 @@ Mano-P is based on the following research work: } ``` -### Academic Collaboration +### 学术合作 -We welcome collaboration with academia: +我们欢迎学术界的合作研究： -- 🔬 **Dataset Contribution**: Provide new GUI task datasets -- 🤝 **Joint Research**: Collaborate on edge deployment, quantization optimization, GUI understanding, etc. -- 📚 **Benchmarking**: Test Mano-P on new evaluation sets +- 🔬 **数据集贡献**：提供新的GUI任务数据集 +- 🤝 **联合研究**：在端侧部署、量化优化、GUI理解等方向合作 +- 📚 **基准测试**：在新的评测集上测试Mano-P -For academic collaboration inquiries, please contact: model@mininglamp.com +如有学术合作意向，请联系：model@mininglamp.com --- -## ❓ FAQ +## ❓ 常见问题
-🤖 What is Mano-P? +🤖 Mano-P 是什么？
-Mano-P is an **open-source GUI-VLA (Vision-Language-Action) agent** designed to run locally on Apple Silicon edge devices. It uses **pure visual understanding** to automate desktop GUI operations across platforms. +Mano-P 是一个**开源的 GUI-VLA（Vision-Language-Action）智能体**，设计用于在苹果芯片边缘设备上本地运行。它使用**纯视觉理解**来跨平台自动化桌面 GUI 操作。
-⚖️ How does Mano-P compare to Claude Computer Use? +⚖️ Mano-P 与 Claude Computer Use 相比如何？
-**Performance Comparison:** +**性能对比：** -- OSWorld (all models): Claude Sonnet 4.6 **72.1%** vs Mano-CUA 1.1 **58.2%** -- WebRetriever Protocol I: Mano-CUA 1.1 **41.7 NavEval** vs Claude 4.5 Computer Use **31.3** +- OSWorld（所有模型）：Claude Sonnet 4.6 **72.1%** vs Mano-CUA 1.1 **58.2%** +- WebRetriever Protocol I：Mano-CUA 1.1 **41.7 NavEval** vs Claude 4.5 Computer Use **31.3** -**Key Difference:** +**核心差异：** -- ✅ Mano-P **runs entirely on-device**, no data leaves the machine -- ⚠️ Claude Computer Use requires cloud API calls +- ✅ Mano-P **完全在设备上运行**，数据不离开设备 +- ⚠️ Claude Computer Use 需要云端 API 调用 -**Use Case:** Mano-P is particularly suitable for **high-security scenarios**. +**适用场景：** Mano-P 特别适合**高安全性要求**的场景。
-🔌 Can Mano-P run without internet? +🔌 Mano-P 可以离线运行吗？
-**Yes!** In local mode, all model inference runs on the Apple M4 device. ✅ **No screenshots or task descriptions are sent to external servers.** +**可以！** 在本地模式下，所有模型推理都在 Apple M4 设备上运行。✅ **不会向外部服务器发送任何截图或任务描述。**
-💻 What hardware do I need? +💻 需要什么硬件配置？
-**Minimum Requirements:** +**最低要求：** -- Mac mini or MacBook -- Apple M4 chip -- 32GB RAM +- Mac mini 或 MacBook +- Apple M4 芯片 +- 32GB 内存 -**Alternative:** +**替代方案：** -- Any Mac + Mano-P computing stick (connected via USB 4.0+) +- 任何 Mac + Mano-P 算力棒（通过 USB 4.0+ 连接） -📌 We plan to support more devices in the future. +📌 我们计划在未来支持更多设备。
-📦 How do I install Mano-P? +📦 如何安装 Mano-P？
-**CLI Tool:** +**CLI 工具形式：** ```bash brew tap Mininglamp-AI/tap && brew install mano-cua ``` -**OpenClaw/Claude Code Skill:** -See [ClawHub - Mano-CUA](https://clawhub.ai/hanningwang/mano-cua) +**OpenClaw/Claude Code Skill 形式：** +请参见 [ClawHub - Mano-CUA](https://clawhub.ai/hanningwang/mano-cua)
-🔒 Is my data safe? +🔒 我的数据安全吗？
-**Local Mode:** ✅ All processing happens on-device +**本地模式：** ✅ 所有处理都在设备上进行 -**Cloud Mode:** +**云端模式：** -- ⚠️ Only screenshots and task descriptions sent to `mano.mininglamp.com` -- ✅ No local files, clipboard contents, or credentials accessed +- ⚠️ 仅截图和任务描述发送到 `mano.mininglamp.com` +- ✅ 不访问本地文件、剪贴板内容或凭证 -**Transparency:** Full client is [open-source](https://github.com/Mininglamp-AI/mano-skill) for audit +**透明度：** 完整客户端[开源](https://github.com/Mininglamp-AI/mano-skill)可供审计
--- -## 🤝 Contribution Guidelines +## 🤝 贡献指南 -We welcome community contributions! If you want to contribute to the project: +我们欢迎社区贡献！如果你想为项目做出贡献： -1. Fork this repository -2. Create your feature branch (`git checkout -b feature/AmazingFeature`) -3. Commit your changes (`git commit -m 'Add some AmazingFeature'`) -4. Push to the branch (`git push origin feature/AmazingFeature`) -5. Open a Pull Request +1. Fork 本仓库 +2. 创建你的特性分支 (`git checkout -b feature/AmazingFeature`) +3. 提交你的更改 (`git commit -m 'Add some AmazingFeature'`) +4. 推送到分支 (`git push origin feature/AmazingFeature`) +5. 开启一个 Pull Request -### Contribution Areas +### 贡献方向 -- 🐛 Bug fixes and issue reporting -- 📝 Documentation improvements and translations -- 💡 New feature suggestions and implementations -- 🧪 Test cases and benchmarking -- 🎨 Application scenarios and demo contributions +- 🐛 Bug修复和问题报告 +- 📝 文档改进和翻译 +- 💡 新功能建议和实现 +- 🧪 测试用例和基准测试 +- 🎨 应用场景和Demo贡献 --- -## 📄 License +## 📄 开源协议 -This project is licensed under the [Apache License 2.0](LICENSE). +本项目采用 [Apache License 2.0](LICENSE) 开源协议。 -**License Highlights:** +**协议要点：** -- ✅ Commercial use -- ✅ Modification and distribution -- ✅ Patent grant -- ⚠️ Must retain copyright notice -- ⚠️ Must state changes +- ✅ 商业使用 +- ✅ 修改和分发 +- ✅ 专利授权 +- ⚠️ 需保留版权声明 +- ⚠️ 需说明修改内容 --- -## 📮 Contact +## 📮 联系方式 - + -- 📧 Email: model@mininglamp.com -- 🏠 Website: [https://github.com/Mininglamp-AI/Mano-P](https://github.com/Mininglamp-AI/Mano-P) -- 💬 Community: (To be added) +- 📧 邮箱: model@mininglamp.com +- 🏠 官网: [https://github.com/Mininglamp-AI/Mano-P](https://github.com/Mininglamp-AI/Mano-P) +- 💬 社区: （待补充） - 🐛 GitHub Issues: [https://github.com/Mininglamp-AI/Mano-P/issues](https://github.com/Mininglamp-AI/Mano-P/issues) --- -## 🙏 Acknowledgments +## 🙏 致谢 -Thanks to all developers and researchers who contributed to this project. +感谢所有为本项目做出贡献的开发者和研究者。 -**Special Thanks:** +**特别感谢：** -- Mano project team for providing the technical foundation -- DeepMiner platform for deep integration support -- Edge computing hardware partners -- Open source community contributors +- Mano项目团队提供的技术基础 +- DeepMiner平台的深度集成支持 +- 端侧算力硬件合作伙伴 +- 开源社区的贡献者们 ---

- Built with ❤️ by the Mano-P Team + 由Mano-P团队用 ❤️ 打造