Terminal version of Langfuse (Langfuse CLI)

[Shahin-rmz]

Describe the feature or potential improvement

Hello,
Thanks for providing observation and engineering tool for Opensource world.
Just as terminal junky person, wanted to know, if we have any terminal version of Langfuse or something?
Thanks

Additional information

No response

[marcklingen]

Interesting idea, there is no cli yet, let's track interest for this here

[Shahin-rmz]

Thanks for the reply,
I mean as the project has API architecture and it has front end Back end structure, it would be not that hard to implement CLI.
Other than that, if we want to have interactive Tui window, some libraries like textualise are very helpful !
I hope I can do something for the project.

[vicentereig]

I've put together a quick cli over here that I use with codex an claude. Happy hacking.

https://github.com/vicentereig/lf-cli

⏺ Looking at those 2 traces more closely:

  Trace 1: "optimization.miprov2_compile"
  - This is a DSPy MIPROv2 optimization run from your dspy-ruby project
  - It's optimizing a ReAct agent for HotPotQA question answering
  - Configuration:
    - 10 training examples, 5 validation examples
    - 6 trials in "greedy" mode (explores different prompts/examples)
    - Light mode (faster but less thorough)
  - 152 observations over 82.9 seconds
  - Cost: $0.0062
  - This is the optimizer trying different combinations of prompts and few-shot examples to improve the agent

  Trace 2: "evaluation.batch"
  - This is a batch evaluation of the DSPy::ReAct agent
  - Ran on 5 examples with 3 parallel threads
  - Had a metric to measure performance
  - 31 observations over 18 seconds
  - Cost: $0.0010
  - This likely ran before the optimization to get a baseline score

  So these traces tell the story: Evaluate baseline ReAct performance → Optimize with MIPROv2 → (presumably) Re-evaluate to see improvement. This is a
  typical DSPy optimization workflow for improving a question-answering agent.

[Tar-ive]

Thank you for creating this, still need to put this into actual deployment but looks really well documented, and usable.

[vicentereig]

good stuff - I've been using it from codex/claude while running dpsy.rb optimizations and I'm pretty happy. Hoping it works for you too!

[jkryanchou]

Awesome, and now langchain already provides the langsmith-fetch cli tools which could be invoked with Claude Code and deepagents.

I suggested that the langfuse could do the same thing order to provide an methodology to level up the deepagents.