Langfuse v3: Prompts not getting linked to generations in UI when using custom prompt loader

[abhisheksinghpibit]

Describe your question

I’m using Langfuse v3 with a custom prompt management layer (supporting both local files and Langfuse prompts).

The issue is that prompts are not getting linked to generations in the UI, even though I am fetching prompts via langfuse_client.get_prompt().

1. Prompt configuration
   PROMPT_CONFIG = {
   "email_subject_parsing": {
   "local_file": "prompts/email_subject_parsing.txt",
   "lf_key": "email_subject_parsing",
   "system_messages_lf": lambda prompt: prompt,
   }
   }

2. Fetching prompt from Langfuse
   def build_litellm_prompt_lf(config):
   prompt_obj = langfuse_client.get_prompt(
   config["lf_key"],
   label=label
   )

   Using prompt object here

   system_messages_lf = config"system_messages_lf"

   ❗ Converting to string here

   return system_messages_lf.prompt

3. Building prompt dictionary
   def get_litellm_prompt_dict():
   prompt_dict = {}

   for key, config in litellm_prompt_config.items():
   prompt_dict[key] = build_litellm_prompt_lf(config)

   return prompt_dict

4. Creating LLM messages
   litellm_prompt_dict = get_litellm_prompt_dict()

messages = [
{
"role": "system",
"content": litellm_prompt_dict["email_subject_parsing"]
}
]
5. LLM call (wrapped with Langfuse)
def call(self, **kwargs):
metadata = kwargs.get("metadata", {})

with langfuse.start_as_current_observation(
    as_type="generation",
    name=metadata.get("generation_name", "llm_call")
):
    return self._client.call(**kwargs)

Langfuse Cloud or Self-Hosted?

Self-Hosted

If Self-Hosted

No response

If Langfuse Cloud

No response

SDK and integration versions

No response

Pre-Submission Checklist

• I have checked for existing issues/discussions and consulted Langfuse AI.

[inkeep-ai-assistant[bot]]

The issue is that you're converting the prompt object to a string (.prompt) before passing it to your LLM call(1). This breaks the linkage that Langfuse needs to display the prompt connection in the UI.

When using the Langfuse API directly or converting prompt objects to strings, the UI does not recognize this as a managed prompt link(1). The prompt must be fetched using the SDK's get_prompt method and assigned as the prompt value in the generation context to ensure the linkage is recognized and visualized in the Langfuse UI(1).

To fix this, you need to pass the actual prompt object (not the string content) to your generation. Here's how to properly link prompts:

Python SDK approach:

from langfuse import observe, get_client

langfuse = get_client()

@observe(as_type="generation")
def nested_generation():
    prompt = langfuse.get_prompt("movie-critic")
    langfuse.update_current_generation(
        prompt=prompt,
    )

(2)

For your specific case, instead of extracting just the prompt text with .prompt, you should pass the full prompt object when creating your generation. The SDK handles the required internal structures and metadata, ensuring that the prompt linkage appears in the UI diagrams and trace flows(1).

Additionally, if you're using Langfuse v3.109.0 → 3.139.0, note that there was a known frontend regression where prompt links weren't displayed in generation spans (though they appeared in the Observation view). This was fixed in version 3.140.0(3).

📚 Sources:

1. Prompt Management: API-linked Generations Not Showing Prompt Link
2. Troubleshooting and FAQ
3. bug: Prompt link no longer displayed in generation span after SDK update (3.109.0 → 3.139.0)

---

Have another question? Just tag @inkeep.

[abhisheksinghpibit]

i am currently on langfuse==3.14.5

[CodeCLS]

Hi @abhisheksinghpibit, were the points of Inkeep helpful?

[abhisheksinghpibit]

i tried but it still didn't help