[Bug]: The RagFlow HTTP API for updating the chat assistant cannot update the model.

[PaoGitCode]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

RAGFlow workspace code commit ID

53c59c4

RAGFlow image version

0.18

Other environment information

Actual behavior

When I update the model selection through the Update chat assistant interface: PUT /api/v1/chats/{chat_id}, and provide llm.model_name=deepseek-r1:32b to the interface, it can be successfully saved. However, the assistant session cannot call the model because the @ollama field is not added to my model in the above code, resulting in llm_id=deepseek-r1:32b. If I want llm_id to be saved correctly, I need to set llm.model_name=deepseek-r1:32b@Ollama, but this makes it impossible for me to verify through the source code in the above screenshot.

Expected behavior

The llm.model_name=deepseek-r1:32b I submitted should be processed by adding the @ suffix, or when verifying the query, the @ suffix should be automatically truncated for the query.

Steps to reproduce

Submit a modification to the chat assistant through the "Update chat assistant" method by sending a PUT request to /api/v1/chats/{chat_id}, and update and submit the value of llm.model_name

Additional information

No response

[dosubot]

Based on the code and documentation, the '@ollama' suffix is used as a factory identifier to distinguish between different model providers in RagFlow. When you update the chat assistant via the PUT /api/v1/chats/{chat_id} endpoint and set llm.model_name=deepseek-r1:32b, the system saves this value, but the assistant session cannot call the model unless the '@ollama' suffix is present. This is because the code expects the model name to include the factory identifier for correct lookup and usage during inference. If the suffix is missing, the system may not be able to match the model to the correct provider, resulting in issues during model invocation.

The relevant logic is in the split_model_name_and_factory method, which splits the model name by the '@' character. If the suffix is not present, it returns the model name with a None factory identifier, which can cause the lookup to fail if the factory is required for that model type. The get_api_key method then uses this split to query the database for the model configuration, and if the factory identifier is missing, it may not find the correct entry [source].

The expected behavior is that the system should either:

• Automatically append the '@ollama' suffix when saving or updating the model name if the selected provider is Ollama, or
• Automatically handle the absence of the suffix during lookup by matching the model name and factory appropriately.

Currently, this does not happen automatically, so you must provide the full model name with the '@ollama' suffix for the system to function as expected [source].

I did not find any existing issues or discussions in the repository that address this specific problem with the model update API and '@ollama' suffix handling.

To resolve this, you may need to ensure that the model name you submit always includes the '@ollama' suffix when updating the assistant, or adjust the backend logic to handle the suffix automatically during save and lookup. This would align the behavior with your expectation that the system should process or ignore the suffix as needed.

If you need further guidance on where to make these changes in the codebase, let me know!

To continue talking to Dosu, mention @dosu-bot.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[PaoGitCode]

The current problem is that I am unable to update the value with @ollama through the above interface. This is because the code will query tenant_llm by including the entire string of @ollama, but the llm_name field in that table does not have the @ollama suffix. When the assistant actually calls the model, it does so through the llm_id field in the dialog table. If I were to pass in a model name without the @ollama suffix to pass the verification, the interface would not be able to know and add the suffix when saving the llm_id.

[asiroliu]

@PaoGitCode
This is a known issue that we're currently tracking.

We've now fixed the issues that occurred during dataset creation and updates. Would you be interested in submitting a PR for this?

refer： https://github.com/infiniflow/ragflow/blob/main/api/apps/sdk/dataset.py#L147
https://github.com/infiniflow/ragflow/blob/main/api/utils/api_utils.py#L427

[tardigrde]

I have a similar problem.
When I try

rag_flow.create_dataset(name=dataset_name, embedding_model="bge-m3@Ollama")

I get Unsupported model: <bge-m3@Ollama>, but it's there and set as the default embedding model.

Similarly, when I try

rag_flow.create_chat(name=assistant_name, llm=llm_settings, dataset_ids=dataset_ids)

Exception: model_name google/gemini-2.5-flash-preview@OpenRouter doesn't exist

But this one is set as the default text model.
Do I get it right that these issues come form the same root cause?