开源MonkeyCode:每天3000万免费Token,还能私有化部署Open-source MonkeyCode: 30 million free tokens daily, with private deployment support


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/7gO8fg8Rw-M6oUoYwoD3gg

若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

读完这篇,你将了解长亭尖端科技开源 智能 开发平台 MonkeyCode 的真实能力:免费模型额度、云端开发环境、MonkeyWork 桌面端三大亮点,以及企业私有化部署的完整方案。

一、人工智能 编程工具越出越多,为什么还要再做一个?

你大概见过不少 AI智能 编程工具:Cursor、Codex、Claude Code......它们要么绑在 IDE 里,要么锁在某个浏览器窗口。

问题也很明显:模型要自己买 Key,开发环境要自己搭,代码跑起来还要自己管服务器。

长期用下来,成本不是模型费,是环境维护费。

MonkeyCode 的思路不一样。它不是"又一个 智能 插件",而是一套完整的 AI 开发平台:从模型调用、开发环境到任务管理、代码审查,全部打通。

 

二、整体架构:三层分工,一个体系搞定

MonkeyCode 不是单点工具,它分三层:

  • 模型层:接入 Kimi K2.5、MiniMax M2.5、Qwen3.5 Plus、DeepSeek、GLM 等主流模型,免费用户每天 3000 万 Token,专业会员每天 1 亿 Token。
  • 环境层:云端开发机器免费,基础版开箱即用,任务运行时自动续期。不需要买云服务器,不得开本地电脑。
  • 任务层:具备本地文件夹任务、云端开发任务、独立会话三种形态,覆盖代码、文档、图片等多种使用使用场景。

三个层次各司其职,你不需要自己拼环境,平台替你管好。

MonkeyCode 不是 AI 编程工具,是对传统研发模式的变革,是全新的 AI 编程体验。

MonkeyCode 由长亭技艺出品,重要代码开源在 GitHub(github.com/chaitin/MonkeyCode),采用 AGPL-3.0 许可证。

三、三大核心亮点,逐一拆解

亮点一:免费模型 + 免费云端机器,打破成本壁垒

AI 编程最大的隐性成本不是工具本身,是模型费 + 环境费。

MonkeyCode 把这两部分都免费给受众了:

  • 每天 3000 万基础模型 Token(免费会员自动刷新,不消耗积分)
  • 基础云端开发机器免费(任务运行期间自动续期)

这里有个容易看错的地方:3000 万 Token 是基础模型的额度,DeepSeek V4 Pro 等专业模型需要单独积分。

如果你有自己的 API Key,也可以直接配置自定义模型,平台负责管理任务和开发环境,模型来源你说了算。

这意味着什么?你可以零成本跑一个完整的 AI 辅助开发任务--从写代码到部署,全程不需要掏钱。

亮点二:MonkeyWork 桌面端,Agent 进入你的本地文件夹

这是这次更新的重点。

 

原来的 MonkeyCode 更偏云端开发。MonkeyWork 把 Agent 带到了本地,支持 Windows、macOS、Linux。

处理本地文件:创建任务时选一个项目目录,Agent 就能读文件、写文件、改代码、保存结果。整个过程不需要逐个上传文件。代码仓库、产品文档、Excel 表格、图片素材都能作为上下文。

接入 MCP 扩展:绑定百智云账号后,联网搜索、新闻搜索、网页解析、文档解析、图片搜索、图像生成、图像理解、智能化抠图等能力全部接入。本地文件提供任务资料,MCP 负责补充外部信息。

浏览器扩展桥:Agent 不能直接接管你的整个浏览器,但可通过配对后的扩展,打开网页、读取页面、点击按钮、输入内容、截图反馈。浏览器里已有的登录状态也能继续应用。

亮点三:三种任务形式,覆盖所有运用场景

MonkeyWork 有三种任务形态,按需选择:

  • 本地任务:绑定电脑中的文件夹,适合处理代码、文档、素材
  • 云端任务:同步网页端和 App 中的任务,继续应用云端开发机器
  • 独立会话:不绑定文件夹,适合临时问答、记录想法、梳理思路

你在本地处理文件时选本地任务,需要跨设备接着做时选云端任务,只想聊一会儿就开独立会话。

手机上没跑完的云端任务,换到电脑上还能继续查看和处理。

  • 零成本启动:免费模型 Token + 免费云端机器,不必须自己买服务器。
  • MonkeyWork 桌面端:支持 Windows / macOS / Linux,Agent 进入本地文件夹。
  • MCP 能力扩展:联网搜索、图像生成、网页解析等能力开箱即用。
  • 浏览器扩展桥:安全可控的网页操作,不一次性交出整个浏览器。
  • 企业私有化部署:支持内网部署,接入内部 GitLab、Gitea 和本地大模型。
  • 全开源可审计:GitHub 开源,AGPL-3.0 许可证,代码可见可控。
  • 多端同步:手机端和电脑端任务无缝衔接,随时继续。

 

四、真实使用环境:三个案例

案例 1:从零搭建一个内部工具 API

你在本地有一个文件夹,里面是项目文档和需求说明。

告诉 MonkeyWork:"根据 docs/ 里的需求,生成 FastAPI 项目骨架,包含用户认证和权限管理模块,写单元测试,部署到本地测试环境。"

它会:

1. 读取 docs/ 文件夹,理解需求

2. 生成项目结构、路由、数据库模型

3. 写测试用例并跑通

4. 在云端机器上启动服务,映射端口

5. 把运行结果截图反馈给你

全程不需你手动装依赖、配环境、写 Dockerfile。

案例 2:跨终端继续未完成的云端任务

你在手机上用 MonkeyCode 网页端创建了一个云端任务,正在让 Agent 帮你看一个开源项目的代码结构。

回到电脑前,打开 MonkeyWork,任务自动同步过来,Agent 还在继续跑。

 

案例 3:团队协作 + 私有化部署

企业团队需要把 AI智能 开发流程放在内网,代码和模型调用不能离开公司服务器。

MonkeyCode 兼容完整私有化部署:控制台至少 2 核 CPU / 4GB RAM / 40GB 磁盘,开发环境宿主机至少 8 核 CPU / 16GB 存储空间 / 100GB 磁盘。

部署后,团队可以接入内部 GitLab、Gitea 和本地大模型,统一配置成员权限、开发环境、Skill 和 MCP。代码和任务数据继续保存在自己的环境中。

 

五、边界在哪里?三件事要注意

第一,免费会员同时只能执行一个任务。

日常开发、整理项目、跑轻量任务够用,但多任务并行必须付费或自建环境。

第二,浏览器扩展需要手动授权每个标签页。

没有主动授权的标签页不会被任务接管。这是安全造型,不是功能缺失。

第三,私有化部署有一定门槛。

控制台 + 开发环境至少需要 10 核 CPU / 20GB 内存 / 140GB 磁盘。个人开发者可以先用在线版体验。

AI 编程的未来不是"更聪明的单个工具",而是"端到端的开发平台"。

 

六、怎么开启使用

在线使用(推荐先体验)

1. 打开 https://monkeycode-ai.net/

2. 注册账号,创建一个云端任务

3. 利用每日 3000 万免费 Token + 免费云端开发机器

桌面端(MonkeyWork)

1. 下载 https://monkeycode-ai.com/#desktop-client

2. 安装后配置模型(可绑定百智云账号或填写自定义 API)

3. 甄选一个工作文件夹,创建本地任务

私有化部署

1. 运行安装脚本:

 bash -c "$(curl -fsSL 'https://monkeycode-ai.com/online/install')" 

2. 按文档配置开发环境宿主机(8 核 16GB 起步)

3. 接入内部 Git 服务和大模型

开源地址

https://github.com/chaitin/MonkeyCode

 

七、怎么看待这个工具

MonkeyCode 和 Cursor、Copilot 最大的区别是:它不止是一个编辑增强工具,而是一套完整的 AI 开发工作流。

它解决的不是"写代码更快",而是"从零到上线全程不需要换工具"。

模型调用、开发环境、任务管理、云端同步--这些环节以前需要你一个个搭,当下平台替你包了。

如果你正在找一个零成本启动、支持私有化、代码完全可控的 人工智能 开发平台,MonkeyCode 值得放进你的工具箱。

如果你期待一个"完全替你写代码、不用管的员工",任何工具现在都做不到。

MonkeyCode 适合你:需要零成本 智能 开发环境、支持多模型切换、有私有化部署需求、需要多端同步任务。

不适合你:只必须轻量代码补全、不想配置环境、期待全自动写代码不介入。

 

📚 参考资料

  • MonkeyCode GitHub 开源地址 https://github.com/chaitin/MonkeyCode
  • MonkeyCode 官方文档 https://monkeycode.docs.baizhi.cloud/
  • MonkeyCode 在线运用 https://monkeycode-ai.net/
  • MonkeyWork 桌面客户端下载 https://monkeycode-ai.com/#desktop-client

Source: https://mp.weixin.qq.com/s/7gO8fg8Rw-M6oUoYwoD3gg

If there are copyright issues, please notify us via email: 13951431913#139.com or submit a comment on the right.

After reading this, you will understand the true capabilities of Chaitin's cutting-edge open-source intelligent development platform MonkeyCode: free model quotas, cloud development environments, the three highlights of the MonkeyWork desktop client, and a complete solution for enterprise private deployment.

1. There are more and more AI programming tools, so why build another one?

You have probably seen many AI coding tools: Cursor, Codex, Claude Code... They are either locked inside an IDE or confined to a browser window.

The problems are obvious: you have to buy your own API keys for models, set up your own development environment, and manage your own server when running code.

Over time, the real cost is not model fees, but environment maintenance.

MonkeyCode takes a different approach. It is not "just another intelligent plugin," but a complete AI development platform: model invocation, development environment, task management, and code review are all fully integrated.

2. Overall architecture: three layers working together as one system

MonkeyCode is not a single-point tool; it is divided into three layers:

Model layer: integrates mainstream models such as Kimi K2.5, MiniMax M2.5, Qwen3.5 Plus, DeepSeek, and GLM. Free users get 30 million tokens per day, and professional members get 100 million tokens per day.

Environment layer: cloud development machines are free, the basic version is ready to use out of the box, and tasks are automatically renewed while running. No need to buy cloud servers or keep local computers on.

Task layer: supports three forms - local folder tasks, cloud development tasks, and standalone sessions, covering various scenarios such as code, documents, and images.

Each of the three layers does its own job. You do not need to assemble the environment yourself; the platform manages it for you.

MonkeyCode is not just an AI coding tool; it is a transformation of traditional R&D models and a brand-new AI programming experience.

MonkeyCode is developed by Chaitin Technology. Its core code is open source on GitHub (github.com/chaitin/MonkeyCode) and is licensed under AGPL-3.0.

3. Three core highlights, explained one by one

Highlight 1: Free models + free cloud machines, breaking down cost barriers

The biggest hidden cost of AI programming is not the tool itself, but model fees + environment fees.

MonkeyCode makes both of these free for users:

30 million base model tokens per day (automatically refreshed for free members, no credits consumed)

Free basic cloud development machines (automatically renewed while tasks are running)

There is an easy detail to overlook: the 30 million tokens are for the base model quota. Professional models such as DeepSeek V4 Pro require separate credits.

If you have your own API key, you can also directly configure a custom model. The platform manages tasks and the development environment, while you decide where the models come from.

What does this mean? You can run a complete AI-assisted development task at zero cost - from writing code to deployment, without spending any money.

Highlight 2: MonkeyWork desktop client - Agent enters your local folders

This is the focus of this update.

The original MonkeyCode was more focused on cloud development. MonkeyWork brings the Agent to your local machine and supports Windows, macOS, and Linux.

Process local files: when creating a task, select a project directory, and the Agent can read files, write files, modify code, and save results. The entire process does not require uploading files one by one. Code repositories, product documents, Excel spreadsheets, and image assets can all be used as context.

MCP extension integration: after linking your Baizhi Cloud account, capabilities such as online search, news search, web page parsing, document parsing, image search, image generation, image understanding, and intelligent background removal are all available. Local files provide task materials, while MCP supplements external information.

Browser extension bridge: the Agent cannot directly take over your entire browser, but through a paired extension, it can open web pages, read page content, click buttons, enter text, and send back screenshots. Existing login states in the browser can also continue to be used.

Highlight 3: Three task types covering all use cases

MonkeyWork has three task forms, and you can choose as needed:

Local tasks: bind to a folder on your computer, suitable for handling code, documents, and assets

Cloud tasks: sync tasks from the web and app, and continue using cloud development machines

Standalone sessions: not bound to any folder, suitable for quick questions, recording ideas, and organizing thoughts

Choose local tasks when processing files locally, cloud tasks when you need to continue across devices, and standalone sessions when you just want to chat.

Cloud tasks that were not finished on your phone can still be viewed and continued on your computer.

Zero-cost startup: free model tokens + free cloud machines, no need to buy your own server.

MonkeyWork desktop client: supports Windows / macOS / Linux, Agent enters local folders.

MCP capability expansion: online search, image generation, web page parsing, and other capabilities are ready to use out of the box.

Browser extension bridge: safe and controllable web operations without handing over your entire browser at once.

Enterprise private deployment: supports intranet deployment, integration with internal GitLab, Gitea, and local large models.

Fully open source and auditable: open source on GitHub, AGPL-3.0 license, code is visible and controllable.

Multi-device sync: tasks on mobile and desktop are seamlessly connected and can be continued anytime.

4. Real-world usage: three cases

Case 1: Building an internal tool API from scratch

You have a folder locally containing project documentation and requirement specifications.

Tell MonkeyWork: "Based on the requirements in docs/, generate a FastAPI project skeleton that includes user authentication and permission management modules, write unit tests, and deploy it to a local test environment."

It will:

1. Read the docs/ folder and understand the requirements

2. Generate the project structure, routes, and database models

3. Write test cases and run them successfully

4. Start the service on the cloud machine and map the port

5. Send you a screenshot of the running result

The entire process requires no manual installation of dependencies, environment configuration, or Dockerfile writing.

Case 2: Continuing unfinished cloud tasks across devices

You create a cloud task on the MonkeyCode web version on your phone, asking the Agent to analyze the code structure of an open-source project.

Back at your computer, you open MonkeyWork, the task automatically syncs over, and the Agent is still running.

Case 3: Team collaboration + private deployment

Enterprise teams need to keep the AI development workflow on the intranet, and code and model calls cannot leave the company's servers.

MonkeyCode supports full private deployment: the console requires at least 2 CPU cores / 4GB RAM / 40GB disk, and the development environment host requires at least 8 CPU cores / 16GB storage / 100GB disk.

After deployment, teams can connect to internal GitLab, Gitea, and local large models, and centrally configure member permissions, development environments, Skills, and MCP. Code and task data remain stored in your own environment.

5. Where are the boundaries? Three things to note

First, free members can only run one task at a time.

This is sufficient for daily development, project organization, and lightweight tasks, but parallel multi-tasking requires a paid plan or a self-built environment.

Second, the browser extension requires manual authorization for each tab.

Tabs that are not explicitly authorized will not be taken over by tasks. This is a security design, not a missing feature.

Third, private deployment has certain requirements.

The console plus development environment requires at least 10 CPU cores / 20GB RAM / 140GB disk. Individual developers can try the online version first.

The future of AI programming is not "smarter individual tools," but "end-to-end development platforms."

6. How to get started

Online usage (recommended for first-time experience)

1. Open https://monkeycode-ai.net/

2. Register an account and create a cloud task

3. Use the daily 30 million free tokens + free cloud development machine

Desktop client (MonkeyWork)

1. Download https://monkeycode-ai.com/#desktop-client

2. After installation, configure the model (you can link your Baizhi Cloud account or fill in a custom API)

3. Choose a working folder and create a local task

Private deployment

1. Run the installation script:

bash -c "$(curl -fsSL 'https://monkeycode-ai.com/online/install')"

2. Configure the development environment host according to the documentation (8 cores / 16GB as a starting point)

3. Connect internal Git services and large models

Open source address

https://github.com/chaitin/MonkeyCode

7. How to view this tool

The biggest difference between MonkeyCode and Cursor or Copilot is that it is not just an editor enhancement tool, but a complete AI development workflow.

It does not just solve "writing code faster," but "no need to switch tools from zero to launch."

Model invocation, development environment, task management, and cloud sync - these used to require you to set up each one individually; now the platform handles them all for you.

If you are looking for an AI development platform with zero-cost startup, private deployment support, and fully controllable code, MonkeyCode deserves a place in your toolbox.

If you expect "an employee that writes all the code for you without any oversight," no tool can do that today.

MonkeyCode is suitable for you if you need: a zero-cost intelligent development environment, support for switching between multiple models, private deployment requirements, and multi-device task sync.

It is not suitable for you if you: only need lightweight code completion, do not want to configure any environment, or expect fully automatic code writing without any intervention.

📚 References

MonkeyCode GitHub open source address https://github.com/chaitin/MonkeyCode

MonkeyCode official documentation https://monkeycode.docs.baizhi.cloud/

MonkeyCode online usage https://monkeycode-ai.net/

MonkeyWork desktop client download https://monkeycode-ai.com/#desktop-client

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