How can I select the chat model and embedding model as the same one?

[Ghjk94522]

when I run a ragflow server on my docker, and I want to set default models as follows:

as shown above, moonshow api has both LLM and embedding features, but when I want choose it as default embedding model, I can find nothing in the menu.

Is it a bug of the ragflow app or just can not select one model as different default models?

looking forward to some help, thanks.

[dosubot[bot]]

Based on the available documentation and issues, RAGFlow requires you to add models in the "Model providers" section and then set them as default in "System model settings" for both LLM (chat) and embedding purposes. However, there is no explicit documentation or issue confirming that you can select the same model (such as Moonshot) for both chat and embedding. The fact that the embedding model dropdown shows "No data" for Moonshot, even though it supports both features, suggests this may be a limitation or oversight in the current version of RAGFlow, rather than a confirmed bug or an intended restriction.

If you want to set default models when running RAGFlow on Docker, you should update the service_conf.yaml.template file before starting the server, specifying your chosen LLM and embedding models. After the server is running, you must use the UI to configure models, as changes to the config file will no longer take effect. For more details, see the Docker setup guide and model API key setup.

If you need to use the same model for both chat and embedding and it does not appear as an option, this is likely a current limitation of the app. There is no documentation or issue confirming support for this scenario or describing it as a bug. If this is critical for your workflow, you may want to monitor future releases or updates for expanded model selection support.

To reply, just mention my name @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[Ghjk94522]

@dosu ok, but I think it’s a common requirement to use the same model as the default model for different functions, as most people might only purchase the API from one company.

[dosubot[bot]]

You're absolutely right—using the same model for both chat and embedding is a common requirement, especially when users want to streamline costs and management with a single provider.

Currently, RAGFlow does allow the same model to be set for multiple functions, but only if that model is explicitly supported for each function. For example, models like Azure-OpenAI, BaiChuan, BaiduYiyan, Bedrock, Cohere, and Gemini are supported for both chat and embedding, so you can select them for both roles in the UI or configuration files. However, the Moonshot model is only listed under chat models and not under embedding models in the codebase and documentation, which is why it does not appear as an option for embedding—even if the provider itself supports both features in their API. This appears to be a limitation in the current version of RAGFlow, not a general restriction on using the same model for multiple functions. You can see the full list of supported models and their capabilities in the documentation "Supported Models".

There are examples in the codebase where the same model (such as DeepseekChat) is used for both chat and embedding, confirming that RAGFlow can support this scenario when the model is registered for both purposes in the app logic "model-card.tsx".

If Moonshot adds explicit support for embedding in a future RAGFlow release, or if the maintainers update the model registry, you should be able to select it for both functions. For now, you may want to consider one of the providers that are already supported for both chat and embedding, or keep an eye on future updates for expanded model selection.

To reply, just mention my name @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}