一年之前推荐过一个开源项目,叫 MaxKB。
那时候它靠开箱即用的 RAG 能力,在 GitHub 上拿了 17k Star,定位是企业的最强大脑。
一年过去,这个开源项目全网累计下载正式突破 100 万次,Star 数突破了21K。

这一年一直在往企业级智能体编排这个方向卷。
可以看一下这一年多的版本日志,每个版本都在打磨智能体编排能力,没再满足于装个知识库、问个问题那个初级阶段。
下面我捡几个最喜欢的,跟你们聊聊。
01
开源项目简介
这个 MaxKB 是让业务人员也能搭 AI 工作流的平台,跟同类产品比,MaxKB最大的特点是简单。
工作流编排全靠拖拉拽,不用写一行代码。
平台还自带工具商店和模板中心,合同审计、数据分析、公文写作、邮件推送这类高频场景,点一下就能套用现成的工具和模板。

大模型接入方面,本地能跑 DeepSeek、千问 Qwen、Llama 这些开源模型。
在线的从 DeepSeek、豆包、Kimi 到 OpenAI、Claude、Gemini,主流的都支持。

文档上传或者让它自动爬取在线文档,剩下的拆分、向量化、RAG 流程全自动。
工作流编排和各种工具调用,比如 Skills、MCP工具在平台上点点就接上了,能搞定复杂场景的自动化,进而拼出复杂的业务智能体。
搭好的 AI 助手或智能体可以零代码嵌入网站、企业微信、微信公众号、钉钉、飞书等,让已有的系统快速长出智能交互能力。
它还开放了API,其他应用可以直接调它做审核、审计、决策。
02
这一年更新了啥
① 应用升级为智能体,模板中心加入
MaxKB 把应用模块直接升级成了智能体。

新加的模板中心对我也特别实用。
官方把合同审核、销售报价、CRM 查询这些高频场景打包成标准套件,点一下就能跑。
从自己搭积木变成直接挑积木,省下的时间够我提前半小时收工。
② Python 代码自动生成
如果想增加这个系统的灵活性,有一些节点可以通过脚本的方式来自定义。
但是对于不会写 Python 的用户。
可以直接用 Python 代码自动生成功能,你只需要描述意图,自动生成一段能跑的 Python 片段。
重复又无聊的事让工具干,人腾出手去做更有创造力的事。
③ 长期记忆
MaxKB 智能体长期记忆的原理不复杂:系统按对话轮次或时间周期,自动提炼你的偏好、背景、目标。
通过 {memory} 变量注入上下文。
人机交互从机械问答变成了有连续性的协作。
03
安装与升级
MaxKB 装起来很方便。还是那套最简单的 Docker 启动方式。
Linux 操作系统:
docker run -d --name=maxkb --restart=always -p 8080:8080 \-v ~/maxkb:/var/lib/postgresql/data \-v ~/python-packages:/opt/maxkb/app/sandbox/python-packages \registry.fit2cloud.com/maxkb/maxkb:v1
Windows 操作系统:
docker run -d --name=maxkb --restart=always -p 8080:8080 \-v C:/maxkb:/var/lib/postgresql/data \-v C:/python-packages:/opt/maxkb/app/sandbox/python-packages \registry.fit2cloud.com/maxkb/maxkb:v1
部署完,浏览器访问http://你的服务器IP:8080。
用户名:admin
密码:MaxKB@123..(装完一定要记得改密码)
官方还准备了一系列教学视频,可以看看。
教学视频:https://space.bilibili.com/1538710292
04
I recommended an open-source project called MaxKB a year ago.
At that time, it gained 17K stars on GitHub for its out-of-the-box RAG capabilities, positioning itself as the most powerful brain for enterprises.
After a year, the total downloads of this open-source project have officially surpassed 1 million, and the number of stars has exceeded 21K.
Over this past year, the focus has been on pushing the envelope for enterprise-level agent orchestration.
Looking at the release notes from the past year, every version has been fine-tuning the agent orchestration capabilities, moving beyond the initial stage of simply setting up a knowledge base and asking questions.
Below, I'll share a few of my favorite features and talk about them with you.
01
Open-Source Project Introduction
MaxKB is a platform that allows business personnel to build AI workflows without needing to write any code. Compared to similar products, MaxKB's biggest advantage is its simplicity.
The workflow orchestration relies entirely on drag-and-drop, requiring no programming knowledge.
The platform also comes with a built-in tool store and template center. For high-frequency scenarios like contract auditing, data analysis, official document writing, and email notifications, you can apply ready-made tools and templates with a single click.
In terms of large model integration, it can run local open-source models like DeepSeek, Qwen, and Llama.
For online services, it supports a wide range, including DeepSeek, Doubao, Kimi, OpenAI, Claude, and Gemini.
You can upload documents or let the system automatically crawl online documents; the subsequent splitting, vectorization, and RAG processes are all automated.
Orchestrating workflows and calling various tools, such as Skills and MCP tools, is done directly within the platform. This allows for the automation of complex scenarios and the construction of sophisticated business agents.
The built AI assistants or agents can be seamlessly integrated into websites, enterprise WeChat, public accounts, DingTalk, and Lark without any coding, quickly adding intelligent interaction capabilities to existing systems.
It also offers an API interface, allowing other applications to directly call its auditing, review, and decision-making services.
02
What's New This Year
① Application Upgraded to Agent, Template Center Added
MaxKB has upgraded the application module to a full-fledged agent.
The newly added template center is particularly useful for me.
The official team has packaged common scenarios like contract review, sales quoting, and CRM queries into standard kits that can be run with just a click.
This shift from building blocks from scratch to picking pre-made blocks has saved me a significant amount of time.
② Automatic Python Code Generation
If you want to increase the system's flexibility, some nodes can be customized with scripts.
However, for users who don't know how to write Python, there is a feature for automatic Python code generation. You simply describe your intent, and the system will generate a functional Python snippet.
This allows you to delegate repetitive, low-value tasks to the tool, freeing up your time for more creative work.
The long-term memory feature in MaxKB works by the system automatically extracting your preferences, background, and goals based on dialogue turns or time periods.
This information is then injected into the context via the {memory} variable.
This transforms the human-machine interaction from mechanical Q&A to a continuous, collaborative exchange.
Installation and Updates
Installing MaxKB is very convenient. It uses the standard Docker startup method.
For Linux operating systems:
For Windows operating systems:
After deployment, access the browser at http://your-server-ip:8080.
The default username is: admin
The default password is: MaxKB@123.. (Remember to change the password after installation for security reasons.)
The official website also provides a series of tutorial videos to get you started.
04
04






