I'd like to recommend 6 AI knowledge bases and open-source projects that I've been keeping an eye on.

After being an AI product manager for a long time, in this era of information explosion, there is not only a daily influx of technological information but also a variety of interesting projects emerging from global developers. These projects include large models, agent applications, and even some algorithms.

So, many product managers have adopted a laid - back attitude. They only need to complete their work and avoid learning new things whenever possible.

I once wrote an article stating that excellent product managers must learn about GitHub. Especially if it's 2025 and you still don't have a GitHub account, you can be directly labeled as a product manager who is criticized by programmers or R & D teams, and you're also wasting development resources.

After that article, some friends left messages asking me to recommend some open - source projects that can be used in work. I also introduce these open - source tools and algorithms in each session of the AI Product Manager Training Camp.

In this article today, I'll share the open - source projects I've been keeping an eye on and even using in my team, along with their use cases and the problems they solve. If you learn according to these categories, you won't feel as much pressure.

1. DIFY - like Open - source Projects: AI Model Management Tools

Currently, Dify - like open - source projects mainly focus on model management. COZE, which was recently open - sourced, also falls into this category. They all support managing multiple AI models and debugging them.

By configuring task flows or agents, you can build the AI scenarios you need. Whether it's for diagnosis or knowledge bases, these open - source projects can get the job done.

When development resources are limited and the team doesn't know how to adjust and configure AI models, these tools can quickly manage AI models through a visual interface, reducing R & D time.

Open - source address: https://github.com/langgenius/dify

2. Manus - like Open - source Projects: Automated Production

At first, I mentioned OpenManus, which is the counterpart of Manus. Now, there's Joyagent open - sourced by JD. These open - source tools are agent applications based on large models. The latest version of Manus can now support thousands of agent applications running in parallel. Although some people complain that the official website's example, which only does simple tasks, uses thousands of agents, and the token consumption behind it doesn't match the application's value.

However, it's undeniable that the automated AI agent workflow is one of the important directions that AI product managers focus on. Many AI product managers don't have the opportunity to work on large AI models. Their companies can only provide resources to call cloud APIs, so most of them focus on AI application development. When it comes to application development, Joyagent is at the forefront. After all, it's the state - of - the - art in open - source, surpassing OpenMANUS.

Open - source address: https://github.com/jd-opensource/joyagent-jdgenie

3. Open - source Projects for Development and Design

Besides the well - known Cursor, some time ago, I shared an open - source project called Onlook for product managers to do UI design and front - end coding. It can automatically generate React front - end style coding with one click. There are also projects like figmamaker, which are all open - source "coding tools" for product managers. Product managers can complete front - end coding without writing code.

Other open - source tools like super design are also worth paying attention to.

Open - source project address: https://github.com/superdesigndev/superdesign

4. Open - source AI Models

There are a large number of open - source models on Hugging Face. Due to limited team resources, I mainly focus on several top - ranked open - source models, namely OpenAI's OSS 120B, DEEPSEEK 671B, KIMI2, and Qwen3 from Alibaba Tongyi Qianwen.

Since the video memory used by a single one of these open - source models reaches about 100GB, even with technologies like ktransoformer that shift video memory operations to memory and algorithms, it's still difficult to guarantee the token speed.

Downloading and deploying open - source models also takes time. So, technical teams often choose Dify - like tools to manage these open - source models uniformly with Ollma to avoid interruptions in the business use of their AI applications.

Open - source model address: https://huggingface.co/

5. Open - source Algorithms: Voice, Digital Human Technology, Video, and Image Technology

In addition to open - source models, I'm also paying attention to open - source voice technologies such as TTS and ASR. These open - source technologies are essential for digital human conversations and can't be replaced by general large models.

As shown in the following flowchart, you can see the value and significance of voice and digital human technology.

The goal of open - source voice technology is to find voices that can mimic human emotional communication as much as possible, making it hard to tell they're artificial. Digital humans should look as real as possible. These belong to the deepfake technology category, and it's convenient to find open - source algorithms for integration.

6. Focus on Only 2 Open - source Licenses

For all the above open - source projects, we only focus on the MIT and Apache 2.0 licenses, which allow commercial use and modification, and don't require public disclosure after modification.

This way, we can ensure the commercialization of our products while also having R & D value, rather than just playing around.

That's all for today's sharing. I hope you can also keep an eye on these directions like me because they're constantly evolving. If you archive these types of open - source projects, you'll find that there's less learning pressure and anxiety.

This article is from the WeChat official account "Kevin's Little Bits of Changing the World" (ID: Kevingbsjddd). The author is Kevin's Stories. It is published by 36Kr with authorization.