There are so many products in the space right now, I hesitate to say “best” because I have only scratched the surface. It’s the best that I’ve tried. The main things it has going for it:
- It’s open source
- It has the right level of abstraction IMO. A lot of products in the space are over engineered, which make them brittle and a pain to work with. Langfuse is focused enough that it does what it needs to well, but is flexible enough to use with a lot of other tools.
- The developers seem competent. Their roadmap looks solid and they develop very quickly for a team their size.
- It works. I can’t tell you the number of times I’ve tried to do something with Langchain products where it just breaks doing simple things. Lots of bugs, bad documentation. Langfuse is better in this regard.
- It’s cheaper (at least it is cheaper than LangSmith). LangSmith charges you for seats and per trace. They charge you even more if you want to keep a trace for more than 14 days, and even then you can only keep traces for a max of 400 days, even if you are hosting on your own servers. Langfuse charges you for seats and that’s it. If you are self-hosting you can do what you want with your data.
There are possibly other platforms that are similar to Langfuse as well, but it’s the best I’ve encountered so far.
For other open-source alternatives, check out https://github.com/agenta-ai/agenta (I am a maintainer), we're more focused on evals and collaboration between subject matter experts and engineers (friendlier UI, powerful prompt engineering flow (playground, eval from UI, etc..)).