Comparing Chinese AI agent and workflow platforms
Machine-assisted translation of the archived article. Read the Chinese original.

| Comparing Chinese AI agent platforms, workflow tools and deployment options | ||||||||
| No. | Product | Provider | Product type | Website | Supported model providers | AI workflows | Other features | Notes |
| 1 | coze-cn | ByteDance | Hosted agent platform | https://coze.cn/ | Supports several providers, with ByteDance’s Doubao models as the main option. The international edition is Coze.com. | Yes | Extensive plugin marketplace | |
| 2 | Alibaba Cloud Model Studio (Bailian) | Alibaba | Hosted agent platform | https://bailian.console.aliyun.com/ | Supports multiple models, primarily Alibaba’s Qwen family. The edition reviewed did not offer models from overseas providers. | Yes | Extensive plugin marketplace | |
| 3 | Baidu AI Agents | Baidu | Hosted agent platform | https://agents.baidu.com/ | The edition reviewed supported only Baidu’s own models. | Yes | Extensive plugin marketplace | Baidu requires business verification to create a branded agent. The platform can also help users discover the agent. |
| 5 | Zhipu AI Agents | Zhipu AI | Hosted agent platform | https://bigmodel.cn/console | The edition reviewed supported only Zhipu AI’s models. | No | Extensive plugin marketplace | |
| 6 | Tencent Yuanqi | Tencent | Hosted agent platform | https://yuanqi.tencent.com/ | The edition reviewed supported only Tencent’s Hunyuan models. | Yes | Extensive plugin marketplace | |
| 7 | Fastgpt,dify,maxkb | The article reports that Alibaba has invested in both Dify and FastGPT. | Self-hosted agent platform | Supports models from Chinese and international providers. | Yes | Extensive plugin marketplace | Connects to databases such as MySQL. The author suggests considering self-hosting when deploying more than ten agents. | |
| 8 | Readers are welcome to suggest additional platforms. | |||||||
On a given platform, the author sees the knowledge base as the element that distinguishes one agent from another.
Common configurations are easier to replicate than proprietary knowledge.
Teams can reuse a model, plugin, prompt or even an entire workflow node.
A knowledge base can travel with the team. Once assembled, its contents can be moved to another agent platform.
Results will still depend on the underlying language model and the embedding model.
Why create a branded agent on a hosted platform?
Hosted platforms can help people discover a brand. The author compares adding a well-curated brand knowledge base to creating a Baidu Baike entry, and speculates that future models might recognise a brand through material made available to them.
Could a platform use your knowledge base to train its next model?
Read each provider’s privacy and intellectual property terms before choosing a platform. These terms affect how your organisation’s knowledge can be used.
The author cites OpenAI’s API terms as an example of a provider that does not use submitted API data for training by default.
Can you call the agent through an API?
Most platforms in the comparison offered API access. Teams should factor the choice of model and its API usage fees into deployment costs.
How should you choose an agent platform?
For teams that need international models, the author suggests a self-hosted platform such as FastGPT or Dify.
Platforms vary in engineering maturity and commercial adoption. Consider which models they support and whether they can help users discover your agent.
If you need more than ten agents, the author suggests running your own platform to manage AI content and resources in one place.
