Create traces for playground executions

[fwang2002]

When I run in playground, the model successfully returned output. But there is no trace.
I use a openai compatible llm.

[inkeep-ai-assistant[bot]]

When you run a prompt in the Langfuse Playground using an OpenAI-compatible LLM, the output is returned, but a trace is not always created. The Playground is primarily for prompt testing and iteration, and does not automatically generate traces for every run(1)(2). This is expected behavior: traces are created when prompts are executed through your application with tracing enabled (such as via the Langfuse SDK or OpenAI integration), but not for all Playground interactions(1)(2).

If you need to generate a trace for Playground runs, you must execute the prompt through your instrumented application, not just the Playground UI(1)(2).

📚 Sources:

1. LLM playground
2. LLM playground - Langfuse

---

Have another question? Just tag @inkeep.

[clemra]

We don't currently log traces when executing in the playground.

[danilotpnta]

just check it out Langwatch and they do, I want to see latency and cost from my different prompts, also how can I check my logs for this, it seems like we should allow for logging as other providers do

[modecolems]

Hi,

Indeed, It could be very helpful to have traces from the playground (as in the Langsmith playground).
When comparing a prompt over different LLM configurations, it could be great to have an idea of the resulting cost, the latency, etc.