Unboxing Coze's Open-Source Version: Agent's Three Core Components Fully Revealed, Attracts 9K Stars in 48 Hours

It still adopts the most permissive Apache 2.0 open-source license.

There is now an open - source one - stop solution for Agent development!

Recently, two sub - products of Coze have been officially open - sourced: Coze Studio and Coze Loop.

Just over a weekend, the two projects have gained 9K Stars!

Together with the previously open - sourced development framework Eino, Coze has open - sourced the entire process of Agent development, evaluation, and operation and maintenance, which is truly a one - stop solution.

It's obvious to all how popular Agent has been this year.

Whether it's the emergence of various popular Agents or the major manufacturers' successive release of MCP protocol support, all signs indicate one thing: Agent is evolving from a "showy toy" to a real - world application tool.

At this critical moment, it's well - known that the maturity of Agent can't rely on a single manufacturer alone.

Open - source is one of the keys to break the deadlock —— By attracting global developers to participate, it can inject strong impetus into the transformation of Agent from a tool to an ecosystem.

So this time, the several products open - sourced by Coze cover the entire lifecycle of Agent development:

Coze Studio: A low - code visual Agent development platform that enables developers to quickly build AI workflows;
Coze Loop: A platform for Prompt development, evaluation, and operation and maintenance, ensuring the stability and optimization space of Agent;
Eino: A unified and abstract AI application orchestration framework that allows Agent to flexibly connect to different models and data sources.

For example, Coze has laid the foundation, and in the future, developers can develop Agents as easily as building with Lego bricks.

More importantly, they adopt the most permissive Apache 2.0 open - source license:

It not only allows commercial use but also includes clear patent licensing terms to provide legal protection for users. Users don't need to open - source their modified code and can use it in a closed - source manner. They only need to retain the original copyright notice, disclaimer, and notice files when distributing.

p.s. There are no additional agreements!

Open - sourcing the entire process of Agent development: development, evaluation, and operation and maintenance

Without further ado, let's unpack the Coze open - source trio.

Coze Studio

In the Agent era, when you have a great idea but are troubled by the high threshold of implementation, Coze Studio comes in handy.

As the name suggests, it is specifically designed to help everyone build Agents more easily.

This open - source platform is comprehensive, providing almost all the core technologies required for Agent development.

To put it simply, from development to deployment, Coze Studio has considered almost everything that developers might need. (It has gained 7.3k Stars just a few days after being open - sourced)

Open - source address of Coze Studio: https://github.com/coze-dev/coze - studio

Of course, having a wide range of tools is one thing, and the most important thing is whether they are easy to use.

In this regard, Coze Studio has some clever designs.

Notably, it has a complete workflow engine. This is the core brain of Agent, containing all Coze's node types and orchestration logic.

Node types can be understood as small functional building blocks that can be dragged into the workflow. For example, the LLM node is used to call large language models to generate content. Orchestration logic refers to the rules and processes of how nodes are connected and executed in what order.

With these two things, developers can develop Agents more easily with just some drag - and - drop actions.

In addition, Coze Studio also supports a "plugin system", which is like giving an "add - on" to Agent.

This time, the platform has open - sourced the definition (how to write a plugin), invocation (how the plugin is called by Agent), and management mechanism (how to manage, update, and debug plugins uniformly) of plugins. Developers can write their own plugins following these guidelines.

Moreover, in addition to various built - in plugins, the platform provides a complete set of capabilities for developers to easily create and integrate any third - party API, and offers a wealth of official open - source plugins as examples.

In a word, everything is designed to more conveniently add various additional capabilities to Agent.

After the Agent is deployed, developers can immediately get a fully - functional and user - friendly operation backend, including full - process functions such as Agent creation, debugging, plugin access, and workflow orchestration.

All the above capabilities are ready - to - use. Developers only need to connect their own model API Key to start building Agents on Coze Studio.

Coze Loop

After we "build" an Agent, the next step is to consider how to keep the Agent running stably.

Generally, before an Agent is officially launched, people often encounter the following problems:

Is my Prompt written correctly?
Why does the model's response always deviate?
There is a bug in the Agent. How can I find the problem?
After the version is updated, which one has better performance, the new or the old?
...

This is where Coze Loop shines —— It can solve all the challenges that an Agent faces throughout its lifecycle, from development, debugging, evaluation, to monitoring. (It has gained 2k Stars just a few days after being open - sourced)

Open - source address of Coze Loop: https://github.com/coze - dev/coze - loop

Let's start with the Prompt development module of Coze Loop. It provides full - process support for developers from writing, debugging, optimizing to version management.

Through the visual Playground, developers can test the output effects of different Prompts in real - time and switch between different large models with one click for horizontal comparison.

Without this function, people would have to jump between different model websites and manually copy and paste Prompts for comparison.

With a more convenient comparison method, we also need to evaluate the quality of the model output from a scientific and reasonable perspective.

The evaluation module of Coze Loop then comes into play —— It is a systematic automatic detection mechanism that can help developers evaluate the response effects of Agent from multiple dimensions such as accuracy, conciseness, and compliance.

OK, after dealing with the input and output, there is still a need for full - process observation.

Coze Loop is still there. It will completely record every processing step from input to output, including key nodes such as Prompt parsing, model invocation, and tool execution.

Moreover, even in case of an exception, it will automatically record the specific situation of the exception, pinpoint the problem area, thus saving the trouble of one - by - one point - by - point troubleshooting.

Development framework Eino

In addition to the above two newly open - sourced products, Coze also open - sourced an Agent development framework based on the Go language —— Eino (pronounced like "I know") in February this year.

So far, it has gained 5.3k Stars on the open - source community GitHub and has a certain influence.

Open - source address of Eino: https://github.com/cloudwego/eino

The reasons for its popularity can be summarized as follows:

1. Unified component abstraction and rich implementations

It breaks down the commonly used functions in AI application development into small modules. These modules have unified definitions, so developers don't need to figure out how to implement each function. And each module has multiple implementation methods, which are ready - to - use, saving the trouble of writing from scratch.

2. Flexible orchestration capabilities

It provides multiple orchestration methods (Graph, Chain, Workflow) to help developers express complex business logic in a simple way. During the orchestration process, it also supports functions such as type checking, concurrent management, and aspect management.

3. Complete stream processing capabilities

According to whether the input and output are streams, Eino provides four interaction modes: invoke (ordinary request - response interaction), stream (streaming data processing), collect (collecting all data within a period of time for processing), and transform (data conversion), and supports automatic conversion, merging, and copying of streams.

4. Powerful toolchain

It provides visual orchestration and debugging tools. Developers can build an AI application with just a few drags.

Coze's open - source initiative will drive Agent to be deployed in more application scenarios

To summarize the relationship between the Coze open - source trio, for example:

Coze Studio is like a construction site for building a house, helping to build the framework; Coze Loop is like a quality inspection center, ensuring that the house is problem - free; and Eino is like a smart home system, making the house smarter and more flexible.

It can be said that Coze has taken the lead in fully implementing a closed - loop Agent infrastructure covering the entire process of development, evaluation, and operation and maintenance in the open - source community.

Since the threshold for developing Agents has been significantly reduced, developers can devote more energy to thinking about business logic and scenario innovation in the future.

Therefore, in the future, Agents are likely to be deployed in more industries and scenarios faster than we expected.

In practice, Coze's open - source initiative is particularly suitable for the following typical application scenarios:

Firstly, enterprise internal automation. Common internal processes such as work - order processing, process approval, and knowledge Q&A can now build intelligent assistants through the open - source version of Coze to achieve automatic response and efficient processing, thereby further improving operational efficiency.

Secondly, for small and medium - sized teams and entrepreneurs. Even if they lack complete engineering resources, they can quickly build their own intelligent assistants, chatbots, or business automation processes with the open - source version of Coze, and focus more on business logic and product innovation.

In vertical industry scenarios (such as law, medicine, finance, e - commerce, etc.), developers can quickly connect their own knowledge bases, business rules, and model interfaces based on the open - source version of Coze to create customized Agents for industry needs.

Finally, the open - source version of Coze is also suitable for education, scientific research, and open - source projects. Whether it's for teaching experiments, academic research, or exploring new paradigms in the open - source community, developers can use Coze as a basic tool to flexibly test key capabilities such as model scheduling and inference chain construction.

In a word, Coze's open - source initiative makes it possible to build Agents that are "accessible to all and applicable in diverse scenarios".

On the verge of Agent's explosion, Coze's open - source initiative aims at more application deployments

Of course, the above are just the results of open - source. Before that, the more fundamental question is:

Why does Coze choose to open - source?

In summary, it's just one sentence —— On the verge of Agent's explosion, Coze aims at more application deployments through open - source.

As Agent develops, the entire tech circle can feel its "heat", both on the surface and behind the scenes.

On the one hand, the underlying technologies supporting Agent's booming development are continuously and rapidly evolving. In March this year, Nature reported a new discovery from the non - profit research institution METR —— the "Moore's Law of Agents", that is, the capabilities of Agents are significantly improving at a rate of "doubling every 7 months".

On the other hand, the ecosystem in the spotlight is also heating up rapidly. Globally, both tech giants and startups are competing to enter the market, and various Agent applications are everywhere, which is dazzling.

However, behind this, the large - scale popularization of Agents still faces two major practical obstacles.

One is the unstable user experience. Currently, most Agents are limited by the technological maturity. When performing real tasks, they are almost like a "lottery draw", often being criticized by netizens as "useless once used", let alone forming user stickiness.

Secondly, the development threshold is still too high. From building workflows, accessing multi - modal capabilities to evaluation and optimization, almost every step requires developers to invest a large amount of engineering resources and technical accumulation. Non - professional teams can't enter the market quickly.

When the above "hot" and "cold" factors collide, the industry has gradually realized that:

To promote Agent from being "fun" to being "useful", not only do platform manufacturers need to make individual efforts, but also global developers need to jointly create and build, and promote the prosperity of the technological ecosystem through open - source and co - construction.