GitHub 上 4000 星星的桌面端 Agent ,每天 3000 万免费 Token。A desktop agent with 4000 stars on GitHub, offering 30 million free tokens per day.


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/Ldjf09vZ9xuyRUaNJ-AF_w
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

之前我提到过 MonkeyCode(了解更多) 这个开源项目。

这个开源项目自发布以来,持续不断地进行更新。

从最初的 AI 编程平台,逐步扩展到移动端、私有化部署,再到近期推出的 MonkeyWork 桌面端,其应用场景已日益广泛。

GitHub 上 4000 星星的桌面端 Agent ,每天 3000 万免费 Token。GitHub 上 4000 星星的桌面端 Agent ,每天 3000 万免费 Token。

此外,它目前依然非常大方地提供免费资源。

免费用户每天可使用 3000 万基础模型 Token,同时也能免费使用云端的基础开发环境。

用户无需自行购买 API,也无需预先准备服务器,完成注册后即可直接创建任务。

这次我们再来了解一下它最近的更新内容,重点讨论一下刚刚发布的 MonkeyWork。

01

开源项目简介

MonkeyCode(了解更多) 是一款面向企业的 AI 开发平台,目前在 GitHub 上已收获超过 4K 个 Star。

它将开发环境、AI 模型、开发任务以及项目所需资源整合到了同一个平台之中。

你可以在网页中设定一个任务,让 AI 分析代码、编辑文件、执行指令、部署依赖以及运行测试。

任务触发后会启动专属的云端开发环境,并且生成的应用支持在线预览。

个人用户只需打开浏览器即可立即使用,无需预先在本地设备上安装各种工具。

对于团队而言,MonkeyCode(了解更多) 能够集中管理成员账号、模型资源、任务队列和开发环境。

它同时支持 GitHub 等代码托管平台,能够围绕 Issue、PR 和 MR 开展开发流程与代码审查工作。

目前,MonkeyCode(了解更多) 已提供网页版、iOS、Android 及桌面客户端,在电脑上未完成的云端任务,切换到手机后仍可继续查看和处理。

  •  
开源地址:https:

02

免费模型、免费机器

AI 编程工具在长期使用过程中,主要的费用支出通常来自模型使用和运行环境。

MonkeyCode(了解更多) 为免费用户分别准备了这两种资源。

每日可使用 3000 万个基础模型 Token。

MonkeyCode(了解更多) 的基础会员当前无需付费,每日自动提供 3000 万个基础模型 Token。

该配额每日刷新,在配额内调用基础模型不会扣除积分。官方配额说明页面列出的基础档包含 Kimi K2.5、MiniMax M2.5 和 Qwen3.5 Plus。

平台还集成了 DeepSeek、GLM 等模型。其中 DeepSeek V4 Pro 属于专业模型,专业会员每日可使用基础模型和专业模型合计 1 亿 Token。

这里很容易看错。

3000 万 Token 是免费用户的基础模型配额,并非 DeepSeek V4 Pro 的专属免费配额。

如果你拥有自己的 API,也可以在 MonkeyCode(了解更多) 中添加自定义模型。平台会负责管理任务和开发环境,模型来源可根据个人需求选择。

云端基础开发机器可免费使用。

仅提供模型是不够的,AI 完成代码编写后还需进行依赖安装、编译处理、测试验证和程序运行。

MonkeyCode(了解更多) 的云端任务系统依托于真实的物理服务器。当用户创建任务时,平台会自动配置开发环境,AI 便可在其中执行终端指令、启动应用程序服务、映射网络端口并实时预览网页效果。

基础型的云服务器资源可免费使用,且在任务正常执行过程中,开发环境会自动获得续期权限。

这样一来,用户无需专门购置云服务器,也不必让个人电脑保持持续运行状态。只要能通过网页或手机访问系统,任务就能持续执行。

免费会员当前支持同时运行一个任务。对于日常开发工作、项目整理以及执行一些轻量级任务而言,现有功能已经能够满足需求。

03 这次要重点关注的是 MonkeyWork

MonkeyCode(了解更多) 近期推出了 MonkeyWork 桌面应用程序,该程序支持在 Windows、macOS 和 Linux 系统上运行。

MonkeyWork 并没有设立独立的代码仓库,其桌面端的所有源代码都整合在 MonkeyCode(了解更多) 的主项目仓库内。

原先的 MonkeyCode(了解更多) 更注重云端开发场景的应用。

MonkeyWork 将 Agent 带入本地文件夹,并将本地文件、模型、MCP 和浏览器操作整合到同一个任务流程中。

它依然能够理解代码,其应用领域已拓展至文档、表格、图片及其他办公资料。

处理本地文件

在创建本地任务时,首先需要选择一个与任务相关的文件夹并完成授权设置。

MonkeyWork 能够在这个目录中读取现有文件,理解目录的上下文信息,创建新内容,还能修改文件并将结果保存回原处。

代码仓库、机型文档、Excel 表格、图片素材均可作为任务上下文的一部分。

整个过程无需逐一上传资料,任务生成的报告、代码和图片将继续保存在本地目录中。

权限范围取决于你选择的文件夹。因此在创建任务时,建议选择具体的项目目录,避免授予整个磁盘或范围过大的目录权限。

接入模型和 MCP

MonkeyWork 支持两种模型规格配置方式。

你可以通过绑定百智云账号来同步模型广场中的模型,也可以在设置界面中填写自己的模型地址和 API Key。

绑定百智云之后,已启用的 MCP 服务会自动迁移至桌面程序中。

目前平台支持接入网络检索、资讯检索、网页分析、文档分析、图库检索、图像创建、图像解析及智能图像分离等功能模块。

任务资料通过本地文件提供,MCP 负责补充外部数据源与图像处理能力。

模型、MCP 以及 MonkeyCode(了解更多) 云端任务各自沿用对应的配额与积分制度,桌面端不会创建额外的收费机制。

继续操作浏览器

许多任务在生成文件后,仍需通过网页端进行后续处理。

MonkeyWork 内置浏览器插件桥接功能。安装插件并完成配对后,可支持打开网页、读取内容、执行点击、录入文本和截图回传等操作。

浏览器中现有的会话状态可以继续保留使用。

这部分设计并未将整个浏览器完全交付给 Agent 控制使用。

用户需先打开目标网页,再通过 MonkeyCode(了解更多) 浏览器助手将当前标签页授权给任务。未进行显式授权的标签页将不被任务接管。

三种任务模式

MonkeyWork 支持三种任务运行模式。

GitHub 上 4000 星星的桌面端 Agent ,每天 3000 万免费 Token。

本地任务可以将电脑中的文件夹与任务绑定,非常适合处理各类代码、文档及素材。

云端任务能够同步 MonkeyCode(了解更多) 网页端和 App 中的任务数据,并继续使用云端开发机器。会员权益、每日额度和积分消耗均沿用当前登录的 MonkeyCode(了解更多) 账号。

独立会话不绑定任何文件夹,更适合进行临时问答、记录想法或梳理思路。

当你在本地处理文件时,可以选择本地任务;需要跨设备继续工作时,应选择云端任务;若只是想短暂交流,则开启独立会话即可。

04 私有化部署与使用方式

MonkeyCode(了解更多) 既可以直接使用官方在线版本,也完全支持完整的私有化部署。

如果团队的核心代码、关键文档和模型调用必须保留在内网中,可以将 MonkeyCode(了解更多) 控制台和开发环境宿主机部署到自建服务器上。

完成私有化部署后,团队可以接入内部的 GitLab、Gitea 和本地大模型,同时统一管理成员权限、开发环境、Skill 与 MCP。所有代码和任务数据将始终存储在团队自有环境中。

官方给出的最低规格建议是:MonkeyCode(了解更多) 控制台至少准备 2 核 CPU、4GB 内存和 40GB 磁盘;开发环境宿主机至少准备 8 核 CPU、16GB 内存和 100GB 磁盘。

在线安装可以使用官方给出的脚本:

  •  
bash -c ”$(curl -fsSL 'https://monkeycode(了解更多)-ai.com/online/install')”

如果只是个人感受,最快的方式还是先用在线版。

打开网站注册账号,创建一个云端任务,就可以使用每日免费模型额度和平台给出的云端开发机器。

想处理电脑里的文件,再下载 MonkeyWork 桌面客户端。首次使用时规格模型,挑选工作文件夹创建本地任务。

需要操作网页,再按照设置页的提示安装并配对浏览器扩展。

在线使用:https://monkeycode-ai.net/MonkeyWork 下载:https://monkeycode-ai.com/开源地址:https://github.com/chaitin/MonkeyCode(了解更多)使用文档:https://monkeycode.docs.baizhi.cloud/

Source: https://mp.weixin.qq.com/s/Ldjf09vZ9xuyRUaNJ-AF_w If there are any copyright issues, please contact us at 13951431913#139.com or submit a comment on the right.

I previously mentioned the open-source project MonkeyCode (Learn More).

Since its release, this open-source project has undergone continuous updates.

Starting from an AI programming platform, it gradually expanded to mobile devices, private deployments, and recently introduced the MonkeyWork desktop client, significantly broadening its application scenarios.

A desktop agent with 4000 stars on GitHub, offering 30 million free tokens per day.A desktop agent with 4000 stars on GitHub, offering 30 million free tokens per day.

Additionally, it remains very generous in providing free resources.

Free users can utilize 30 million base model tokens daily and have free access to cloud-based development environments.

There is no need to purchase APIs or prepare servers in advance; users can create tasks immediately after registration.

This time, we will delve into its recent updates, with a primary focus on the newly launched MonkeyWork.

01

Open-source Project Overview

MonkeyCode (Learn More) is an AI development platform for enterprises, having garnered over 4K stars on GitHub.

It integrates development environments, AI models, development tasks, and necessary project resources into a single platform.

Users can define a task in the web interface, instructing the AI to analyze code, edit files, execute commands, deploy dependencies, and run tests.

Once a task is triggered, a dedicated cloud development environment is launched, allowing for real-time online previews of the generated application.

Individual users can start working immediately from their browser without installing any local tools.

For teams, MonkeyCode (Learn More) facilitates the centralized management of member accounts, model resources, task queues, and development environments.

It also supports integration with code hosting platforms like GitHub, enabling the orchestration of development workflows and code reviews around issues and pull requests.

Currently, MonkeyCode (Learn More) is available as a web application, iOS and Android apps, and a desktop client. This allows users to continue reviewing and processing cloud tasks on their phones if they leave their computers.

02

Free Models, Free Machines

In long-term usage, the primary costs for AI programming tools typically stem from model utilization and computational resources.

MonkeyCode (Learn More) addresses both of these aspects for free users.

Daily Access to 30 Million Base Model Tokens

Currently, the basic membership tier of MonkeyCode (Learn More) is free, granting users 30 million base model tokens per day.

This quota resets daily, meaning calls to the base model within this limit do not deduct credits. The official quota page lists the base tiers as including Kimi K2.5, MiniMax M2.5, and Qwen3.5 Plus.

The platform also integrates other models such as DeepSeek and GLM. Notably, DeepSeek V4 Pro is classified as a professional model, which professional members can use alongside the base model for a total of 100 million tokens per day.

It is important to note that the 30 million tokens are for the base model only and do not constitute a separate free quota for DeepSeek V4 Pro.

Users who wish to use their own models can add them to MonkeyCode (Learn More) by providing their own API keys. In this scenario, the platform will handle the task management and development environment, while the model source remains under user control.

Free Access to Cloud Development Machines

Providing models alone is insufficient; AI also needs to install dependencies, compile, test, and run programs.

The cloud task system of MonkeyCode (Learn More) is powered by physical servers. When a user creates a task, the platform automatically configures the development environment, allowing the AI to execute terminal commands, launch application services, map ports, and perform real-time web previews.

These basic cloud server resources are provided free of charge. As long as the task is running, the development environment remains open and active.

This eliminates the need for users to purchase separate cloud servers or keep their personal computers running continuously. As long as the web or mobile interface is accessible, the task can continue to execute.

Free members can currently run one task simultaneously. This is sufficient for daily development, project organization, and running lightweight tasks.

03 The Highlight: MonkeyWork

MonkeyCode (Learn More) recently introduced MonkeyWork, a desktop application supported on Windows, macOS, and Linux.

Unlike many spin-offs, the entire source code of MonkeyWork is integrated into the main MonkeyCode (Learn More) repository on GitHub.

The original MonkeyCode (Learn More) was heavily focused on cloud-based development.

MonkeyWork brings the Agent concept closer to the user's local filesystem, integrating local files, models, MCP, and browser operations into a single unified workflow.

While it retains the ability to understand code, its application scope has significantly expanded to include documents, spreadsheets, images, and other office data.

Processing Local Files

At the start of a local task, users must select a relevant folder and grant the necessary permissions.

MonkeyWork can read existing files, understand the contextual environment, create new content, and save the results back to the original directory.

This feature supports a wide range of formats, including code repositories, machine manuals, Excel sheets, and image assets.

The entire process is local, eliminating the need for file uploads. Task reports, code, and images are saved directly to the user's local directory.

Permissions are strictly limited to the chosen folder, so it is advisable to select specific project directories rather than granting full drive access.

Connecting Models and MCP

MonkeyWork offers two methods for configuring model specifications.

Users can link their Baizhi Cloud account to synchronize models from the marketplace, or they can manually enter their own model addresses and API keys in the settings.

Linking to Baizhi Cloud ensures that any enabled MCP services are automatically migrated to the desktop application.

Currently, the platform supports integration with web search, news retrieval, page analysis, document analysis, image libraries, image generation, and intelligent image separation modules.

Task data is provided via local files, while the MCP handles external data sources and image processing.

The usage of models, MCP, and MonkeyCode (Learn More) cloud tasks follows the respective credit and quota systems, meaning the desktop application does not introduce new billing structures.

Continued Browser Operations

Many tasks require further processing via a web interface after the initial execution.

MonkeyWork includes a built-in bridge for browser extensions. By installing the plugin and pairing it, users can instruct the agent to open webpages, read content, perform clicks, enter text, and capture screenshots.

Existing browser session states are preserved during this process.

The integration is non-invasive; the user must first open the target webpage and explicitly authorize the tab to the task. Tabs that are not explicitly authorized will not be controlled by the agent.

Three Task Modes

MonkeyWork supports three distinct task execution modes.

A desktop agent with 4000 stars on GitHub, offering 30 million free tokens per day.

Local Tasks bind the agent to a specific folder on the user's computer, making them ideal for handling code, documents, and various media assets.

A desktop agent with 4000 stars on GitHub, offering 30 million free tokens per day.

Standalone Sessions are ephemeral and do not bind to any folder, serving as a quick way to ask questions, jot down notes, or brainstorm ideas.

When working with local files, users should choose local tasks; for cross-device consistency, cloud tasks are appropriate; for quick interactions, standalone sessions are sufficient.

04 Private Deployment and Getting Started

MonkeyCode (Learn More) is highly flexible, supporting both direct online usage and full private deployment.

If a team requires that critical code, documents, and model invocations remain within their internal network, they can deploy the MonkeyCode (Learn More) console and the development environment host on their own servers.

Once deployed, the system can be integrated with internal services like GitLab, Gitea, and local large-scale models, providing a unified management layer for permissions, environments, skills, and MCP. All data and code remain strictly within the team's infrastructure.

The recommended minimum specifications for deployment are 2 CPU cores, 4GB RAM, and 40GB disk for the control panel, and 8 CPU cores, 16GB RAM, and 100GB disk for the development environment host.

For those interested in testing the service, the quickest method is to sign up for the online version.

By creating a cloud task, users can immediately leverage the free daily model quota and the provided cloud development machine.

For local file manipulation, the MonkeyWork desktop client should be downloaded and configured with the necessary folder permissions.

To utilize the browser features, the corresponding extension must be installed and paired according to the instructions in the settings page.

Web Interface: https://monkeyCode-ai.net/

MonkeyWork Download: https://monkeyCode-ai.com/

Open-source Repository: https://github.com/chaitin/MonkeyCode (Learn More)

User Documentation: https://monkeyCode.docs.baizhi.cloud/

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry