/langchain-fixie-marvin

We compared LangChain, Fixie, and Marvin

Primary LanguagePython

Comparing LangChain, Fixie, and Marvin

Task:

  1. Load the "State of the Union" file locally (or remotely) and store them into a vector db (Chroma)
  2. Ask questions for the document

Analysis

Pros:

  • Most powerful option among the three options
  • You can create your own tool.
  • Supports many third-party library integrations, different language models, tracing, and much more.

Cons:

  • Not much complains, but their example notebooks are sometimes broken.

Pros:

  • You can quickly generate and deploy your custom agent
  • You can add your own tool (e.g., @agent.register_func)

Cons:

  • It seems like you cannot choose how the text should be splitted (e.g., text-splitter in LangChain)
  • More examples will be great (i.e., Chat with PDF)
  • Ability to easily add third-party libraries is missing

Misc:

  • Here is a link to my agent.
  • There are no tools for LLM output evaluation at the time of writing this (Apr 23, 2023). However you could add your own validation/evaluation function by integrating a third-party library like Guardrails or Evals.

Pros:

  • AI functions are super cool. (I didn't touch on it in this repo)
  • You can use the LangChain document loader.
  • Interactive option that runs on a terminal is useful.
  • You can add your custom "plugin"s.

Cons:

  • More examples will be great (e.g., Chat with PDF)
  • Ability to easily add third-party libraries is missing